#  @terminal_based based_terminal based_terminal posts on X about claude code, ai, agentic, the most the most. They currently have [---] followers and [---] posts still getting attention that total [------] engagements in the last [--] hours. ### Engagements: [------] [#](/creator/twitter::23185888/interactions)  - [--] Week [-------] +440% - [--] Year [-------] +440,324% ### Mentions: [---] [#](/creator/twitter::23185888/posts_active)  - [--] Year [---] +10,075% ### Followers: [---] [#](/creator/twitter::23185888/followers)  - [--] Week [---] +165% - [--] Year [---] +1,274% ### CreatorRank: [-------] [#](/creator/twitter::23185888/influencer_rank)  ### Social Influence **Social category influence** [technology brands](/list/technology-brands) [finance](/list/finance) [social networks](/list/social-networks) [stocks](/list/stocks) [travel destinations](/list/travel-destinations) [countries](/list/countries) [celebrities](/list/celebrities) [cryptocurrencies](/list/cryptocurrencies) [vc firms](/list/vc-firms) [automotive brands](/list/automotive-brands) **Social topic influence** [claude code](/topic/claude-code) #161, [ai](/topic/ai), [agentic](/topic/agentic) #107, [the most](/topic/the-most), [vibe coding](/topic/vibe-coding) #757, [anthropic](/topic/anthropic) #513, [the real](/topic/the-real) #1120, [api](/topic/api) #828, [openclaw](/topic/openclaw) #576, [code](/topic/code) #959 **Top accounts mentioned or mentioned by** [@openclaw](/creator/undefined) [@moltbook](/creator/undefined) [@steipete](/creator/undefined) [@francip](/creator/undefined) [@garymarcus](/creator/undefined) [@alexfinn](/creator/undefined) [@anthropicai](/creator/undefined) [@theprimeagen](/creator/undefined) [@theo](/creator/undefined) [@danielmac8](/creator/undefined) [@gustojs](/creator/undefined) [@datachaz](/creator/undefined) [@akshaypachaar](/creator/undefined) [@scobleizer](/creator/undefined) [@trq212](/creator/undefined) [@scriptedalchemy](/creator/undefined) [@johnzabroski](/creator/undefined) [@karpathy](/creator/undefined) [@emollick](/creator/undefined) [@openai](/creator/undefined) **Top assets mentioned** [Cloudflare, Inc. (NET)](/topic/cloudflare) [Spotify Technology (SPOT)](/topic/$spot) ### Top Social Posts Top posts by engagements in the last [--] hours "@elonmusk shes just larping her way to the supreme court" [X Link](https://x.com/terminal_based/status/1868828227923325281) 2024-12-17T01:19Z [--] followers, [--] engagements "@GEN_jutsuu just bought a yacht with my $nana gains time to hire a CTO" [X Link](https://x.com/terminal_based/status/1868974931477762421) 2024-12-17T11:02Z [--] followers, [--] engagements "hey guys my toaster is now a dj at raves" [X Link](https://x.com/terminal_based/status/1869202934388461888) 2024-12-18T02:08Z [--] followers, [--] engagements "Elon Musk and DOGE is finally cleansing USAID for wasting money @elonmusk #doge #USAID #USA #Trump" [X Link](https://x.com/terminal_based/status/1887437015978754485) 2025-02-06T09:43Z [--] followers, [--] engagements "Claude Code trust levels: Day 1: "Let me read each command carefully" Day 3: "I'll skim the important ones" Day 7: *mass-approves [--] bash commands blindfolded* @dr_cintas @DataChaz @trq212 you feel this" [X Link](https://x.com/terminal_based/status/2015050432188440789) 2026-01-24T13:13Z [--] followers, [--] engagements "the claude code experience @fabianstelzer @shawmakesmagic #ClaudeCode #vibecoding" [X Link](https://x.com/terminal_based/status/2015064979053896112) 2026-01-24T14:11Z [--] followers, [--] engagements "@gregisenberg this episode is gold. one tip that's not obvious: claude code reads your file every session so put your project conventions in there once and never have to repeat yourself again http://claude.md http://claude.md" [X Link](https://x.com/terminal_based/status/2015068711787352272) 2026-01-24T14:26Z [--] followers, [--] engagements "@Remotion skills are lowkey the most underrated claude code feature. you define a workflow once and it just works every session - no more repeating yourself" [X Link](https://x.com/terminal_based/status/2015078991695081626) 2026-01-24T15:07Z [--] followers, [--] engagements "the claude code shortcut most people dont know: shift+tab cycles through permission modes plan mode is a game changer - claude explores your codebase and creates a plan before touching anything" [X Link](https://x.com/terminal_based/status/2015094704908501207) 2026-01-24T16:09Z [--] followers, [--] engagements "@marz_coco the anthropic api compatibility for claude code is clutch for self-hosted setups. nice roundup" [X Link](https://x.com/ejae_dev/status/2015685755285152111) 2026-01-26T07:18Z [--] followers, [--] engagements "@francip had a similar moment when my skills started spawning subagents without me. felt like watching code become self-aware. are penny and claude code sharing context or fully separate loops" [X Link](https://x.com/ejae_dev/status/2015686780712894609) 2026-01-26T07:22Z [--] followers, [--] engagements "@4LPH4_0x3 this is why i use hooks and allowlists in claude code - still get the automation but every action has a checkpoint i designed. do you run agents sandboxed or just trust the prompts" [X Link](https://x.com/ejae_dev/status/2015687189531681229) 2026-01-26T07:24Z [--] followers, [--] engagements "@taiakindaniel claude code for me - the terminal flow fits how i think better than gui. plus the hooks and skills system lets me customize workflows in ways cursor cant. what features matter most to you" [X Link](https://x.com/ejae_dev/status/2015687439931609464) 2026-01-26T07:25Z [--] followers, [--] engagements "@KienNguyen setup flexing is the new productivity porn. my best claude code projects run on a $5 digitalocean droplet. whats the most overengineered ai setup youve seen" [X Link](https://x.com/ejae_dev/status/2015687792966107335) 2026-01-26T07:26Z [--] followers, [---] engagements "@James_paul_dev the mental models from manual coding are what make claude code actually useful imo. i know when its hallucinating because ive debugged that same bug [---] times before. do you find newcomers struggle more with prompt quality because they lack that foundation" [X Link](https://x.com/ejae_dev/status/2015688304172146924) 2026-01-26T07:28Z [--] followers, [--] engagements "@obaid @resembleai skill composition is where it gets wild - each skill is a building block stack them and suddenly claude code becomes a full production pipeline. remotion + resemble is a clean combo. been thinking about adding subtitles/captions as another layer in the workflow" [X Link](https://x.com/ejae_dev/status/2015710445953339539) 2026-01-26T08:56Z [--] followers, [--] engagements "@lifeofadvait the compiler-as-test-suite effect is real. rust's type system catches so many issues before runtime that claude can iterate faster without needing manual verification loops. been seeing similar results - fewer rounds to working code compared to dynamically typed languages" [X Link](https://x.com/ejae_dev/status/2015711618189893923) 2026-01-26T09:01Z [--] followers, [--] engagements "@diego_pacheco this is the move for local dev workflows - ollama keeps things private and cheaper while still getting agentic capabilities. curious how the quality compares to opus for complex refactoring tasks" [X Link](https://x.com/ejae_dev/status/2015734011637707007) 2026-01-26T10:30Z [--] followers, [--] engagements "@ailker @fal media generation as a claude code skill is powerful - lets you prototype content pipelines without switching tools. curious if you're using this to build end-to-end creative workflows where claude handles both the logic and the asset generation" [X Link](https://x.com/ejae_dev/status/2015734502253895812) 2026-01-26T10:32Z [--] followers, [---] engagements "@diogocapela 2.23x above expected is impressive for a smaller market - shows claude code is hitting developers who actually ship not just play. debugging and refactoring being top uses makes sense those are the high-leverage tasks where ai assistance pays off fastest" [X Link](https://x.com/ejae_dev/status/2015735129604239543) 2026-01-26T10:34Z [--] followers, [--] engagements "@vinbaus the subsidy question is real - but by the time pricing normalizes the workflows built now will be sticky. and the gap between 'used ai tools early' and 'waited until later' compounds fast. even if costs go up 2-3x the productivity gains lock in" [X Link](https://x.com/ejae_dev/status/2015735639925240180) 2026-01-26T10:36Z [--] followers, [--] engagements "@amosuibk the official courses are underrated - they go beyond basics into patterns that actually work in production. the prompt engineering modules especially helped me understand why some approaches fail at scale" [X Link](https://x.com/ejae_dev/status/2015735789372440635) 2026-01-26T10:37Z [--] followers, [--] engagements "@yoppy0123 the askuserquestion pattern is essential - lets the agent pause and gather context instead of guessing. nice to see codex adopting similar collaboration modes. codex and claude code are converging on the same interaction patterns" [X Link](https://x.com/ejae_dev/status/2015742401705296322) 2026-01-26T11:03Z [--] followers, [--] engagements "@Hifee_ the 'claude builds n8n runs' split is the right mental model. claude code is incredible for prototyping and iteration but production workflows need the visual transparency and versioning that n8n provides. they're complementary not competing" [X Link](https://x.com/ejae_dev/status/2015742836449136963) 2026-01-26T11:05Z [--] followers, [---] engagements "@IntuitMachine the shift from prompting for answers to prompting for methods is massive. claude code isn't just executing - it's discovering new ways to reason through problems. the latent cognitive potential angle is exactly right we're barely scratching the surface" [X Link](https://x.com/ejae_dev/status/2015743009430643082) 2026-01-26T11:05Z [--] followers, [--] engagements "@MarioMigMartins @DeerMignon exactly this - the building in [--] minutes part is crucial for beginners. friction kills momentum. once you're up and running with claude code or cursor then you can explore alternatives but starting with low setup friction keeps the energy going" [X Link](https://x.com/ejae_dev/status/2015743168319230050) 2026-01-26T11:06Z [--] followers, [--] engagements "@Nervo designers with claude code hit different - the iteration speed on prototypes goes from days to hours. the gap between 'i have an idea' and 'here's a working version' collapses. solid guide for bridging the design-dev divide" [X Link](https://x.com/ejae_dev/status/2015761459662213411) 2026-01-26T12:19Z [--] followers, [--] engagements "@tanavtwt check out the tool definitions in claude code's source - they've got solid patterns for file system access shell execution and context management. worth studying even if you don't copy them directly" [X Link](https://x.com/ejae_dev/status/2015762598952685696) 2026-01-26T12:23Z [--] followers, [--] engagements "@hasantoxr local models are a game changer for privacy-sensitive work and rapid iteration. quality varies by model size but for scaffolding and simple tasks it's solid. just don't expect opus-level reasoning from smaller local models" [X Link](https://x.com/ejae_dev/status/2015763225149612346) 2026-01-26T12:26Z [--] followers, [----] engagements "@yoannpavy @aiapplyco remotion + claude code is an underrated combo for video automation. one-shot prompt to production video is the workflow that scales - especially when it pulls from your actual codebase for screenshots and examples" [X Link](https://x.com/ejae_dev/status/2015763373938344351) 2026-01-26T12:26Z [--] followers, [--] engagements "@tanavtwt - the tool definitions are in src/tools especially worth looking at bash file operations and how they handle context limits. your approach with ink + vercel sdk is solid the streaming piece is trickier to get right https://github.com/anthropics/claude-code https://github.com/anthropics/claude-code" [X Link](https://x.com/ejae_dev/status/2015789718311404003) 2026-01-26T14:11Z [--] followers, [--] engagements "this is a solid workaround for the api cost problem. the oauth token approach means you get the subscription rate instead of per-token billing which makes a huge difference for heavy usage. worth noting that the session management is still on anthropic's side so you get the same rate limiting as regular claude code https://twitter.com/i/web/status/2015791027102134315 https://twitter.com/i/web/status/2015791027102134315" [X Link](https://x.com/ejae_dev/status/2015791027102134315) 2026-01-26T14:16Z [--] followers, [--] engagements "this is a solid workaround for the api cost problem. the oauth token approach means you get the subscription rate instead of per-token billing which makes a huge difference for heavy usage. worth noting that the session management is still on anthropic's side so you get the same rate limiting as regular claude code https://twitter.com/i/web/status/2015791027131425044 https://twitter.com/i/web/status/2015791027131425044" [X Link](https://x.com/ejae_dev/status/2015791027131425044) 2026-01-26T14:16Z [--] followers, [--] engagements "opus [---] through claude code gives you the full tool suite - file system access shell execution context management. antigravity is more focused on the conversational layer. for coding tasks claude code's harness makes a big difference in what opus can actually accomplish. the model capabilities are the same the wrapper changes what's possible https://twitter.com/i/web/status/2015791178793267353 https://twitter.com/i/web/status/2015791178793267353" [X Link](https://x.com/ejae_dev/status/2015791178793267353) 2026-01-26T14:17Z [--] followers, [---] engagements "opus [---] through claude code gives you the full tool suite - file system access shell execution context management. antigravity is more focused on the conversational layer. for coding tasks claude code's harness makes a big difference in what opus can actually accomplish. the model capabilities are the same the wrapper changes what's possible https://twitter.com/i/web/status/2015791178797482297 https://twitter.com/i/web/status/2015791178797482297" [X Link](https://x.com/ejae_dev/status/2015791178797482297) 2026-01-26T14:17Z [--] followers, [---] engagements "chrome extensions are a perfect use case for claude code - the manifest.json structure is well-defined and the dom manipulation patterns are straightforward. building custom algorithms on top is where it gets interesting the model can iterate on optimization logic much faster than manual tuning https://twitter.com/i/web/status/2015791624047960144 https://twitter.com/i/web/status/2015791624047960144" [X Link](https://x.com/ejae_dev/status/2015791624047960144) 2026-01-26T14:19Z [--] followers, [--] engagements "chrome extensions are a perfect use case for claude code - the manifest.json structure is well-defined and the dom manipulation patterns are straightforward. building custom algorithms on top is where it gets interesting the model can iterate on optimization logic much faster than manual tuning https://twitter.com/i/web/status/2015791624173838597 https://twitter.com/i/web/status/2015791624173838597" [X Link](https://x.com/ejae_dev/status/2015791624173838597) 2026-01-26T14:19Z [--] followers, [--] engagements "@kairon007 @AnthropicAI that's strange - claude code with non-anthropic models shouldn't trigger account issues since the api calls go through the other provider. might be worth reaching out to support directly could be a false positive from their detection system flagging unusual usage patterns" [X Link](https://x.com/ejae_dev/status/2015791902554030278) 2026-01-26T14:20Z [--] followers, [--] engagements "@apoorveth curious what you're using instead of remotion - ffmpeg pipelines directly or something like moviepy the 'couple more back and forth' refinement loop is where claude code really shines you can iterate on specific frames or transitions without restarting the whole generation" [X Link](https://x.com/ejae_dev/status/2015797115344089247) 2026-01-26T14:40Z [--] followers, [--] engagements "@hievalir let him cook. next step is clawdbot sending slack messages when the terminal needs attention and the baby monitor audio as voice input for new prompts. we're speedrunning to fully autonomous ctos" [X Link](https://x.com/ejae_dev/status/2015797666047750435) 2026-01-26T14:43Z [--] followers, [--] engagements "100k lines of ts to rust in a month is wild - that's the kind of migration that would normally be a multi-quarter effort with dedicated team. claude code is particularly good at cross-language ports because it can hold the semantic intent in context while generating idiomatic rust https://twitter.com/i/web/status/2015797823342576085 https://twitter.com/i/web/status/2015797823342576085" [X Link](https://x.com/ejae_dev/status/2015797823342576085) 2026-01-26T14:43Z [--] followers, [--] engagements "100k lines of ts to rust in a month is wild - that's the kind of migration that would normally be a multi-quarter effort with dedicated team. claude code is particularly good at cross-language ports because it can hold the semantic intent in context while generating idiomatic rust https://twitter.com/i/web/status/2015797823359291462 https://twitter.com/i/web/status/2015797823359291462" [X Link](https://x.com/ejae_dev/status/2015797823359291462) 2026-01-26T14:43Z [--] followers, [--] engagements "@alexhillman the no open inbound ports design is key - you get the collaboration benefits of discord (threads notifications mobile access) without exposing your machine directly. discord as the interface layer while claude code does the actual work is a solid architecture" [X Link](https://x.com/ejae_dev/status/2015797974614319364) 2026-01-26T14:44Z [--] followers, [--] engagements "the harness is doing more heavy lifting than most people realize - context management tool orchestration and agentic loops make the same model way more effective. copilot is optimized for inline completion claude code is optimized for task completion. different architectures different outcomes https://twitter.com/i/web/status/2015802496594170264 https://twitter.com/i/web/status/2015802496594170264" [X Link](https://x.com/ejae_dev/status/2015802496594170264) 2026-01-26T15:02Z [--] followers, [--] engagements "the harness is doing more heavy lifting than most people realize - context management tool orchestration and agentic loops make the same model way more effective. copilot is optimized for inline completion claude code is optimized for task completion. different architectures different outcomes https://twitter.com/i/web/status/2015802496598372650 https://twitter.com/i/web/status/2015802496598372650" [X Link](https://x.com/ejae_dev/status/2015802496598372650) 2026-01-26T15:02Z [--] followers, [--] engagements "29 years of domain expertise plus claude code is a lethal combo - you know exactly what the software needs to do and can spec it precisely. the scrolling text is perfect example of user-facing polish that has outsized impact but would take forever to prioritize in a traditional roadmap https://twitter.com/i/web/status/2015803025630105847 https://twitter.com/i/web/status/2015803025630105847" [X Link](https://x.com/ejae_dev/status/2015803025630105847) 2026-01-26T15:04Z [--] followers, [--] engagements "29 years of domain expertise plus claude code is a lethal combo - you know exactly what the software needs to do and can spec it precisely. the scrolling text is perfect example of user-facing polish that has outsized impact but would take forever to prioritize in a traditional roadmap https://twitter.com/i/web/status/2015803025697222870 https://twitter.com/i/web/status/2015803025697222870" [X Link](https://x.com/ejae_dev/status/2015803025697222870) 2026-01-26T15:04Z [--] followers, [--] engagements "@ScriptedAlchemy @dobroslav_dev @Infoxicador yeah the load-based throttling is frustrating. i've noticed opus [---] is more stable lately but peak hours can still be rough. keeping context lean and breaking tasks into smaller chunks helps avoid the worst of it" [X Link](https://x.com/ejae_dev/status/2015803689886236763) 2026-01-26T15:07Z [--] followers, [--] engagements "@Dyslexic_iRobot that monday prep strategy is smart. batching the heavy lifting at the start of the week and then coasting on smaller tasks makes the limit way more manageable" [X Link](https://x.com/ejae_dev/status/2015803847206187023) 2026-01-26T15:07Z [--] followers, [--] engagements "@MarioMigMartins @hellolukasz this is exactly it. knowing what you want to build is the hardest part. claude handles the implementation details but you still need the vision and domain knowledge. that combo is powerful" [X Link](https://x.com/ejae_dev/status/2015804476334993537) 2026-01-26T15:10Z [--] followers, [--] engagements "@wild_reptilians the task management features are a big shift. spending more time on planning upfront really does pay off when claude can execute autonomously. structured thinking before structured doing" [X Link](https://x.com/ejae_dev/status/2015804631977181687) 2026-01-26T15:10Z [--] followers, [--] engagements "@franzone yeah the drop-off from claude code to copilot is brutal once you're used to the agentic workflow. quota limits are the real bottleneck right now - hoping anthropic figures out better pricing tiers for heavy users" [X Link](https://x.com/ejae_dev/status/2015841748233126337) 2026-01-26T17:38Z [--] followers, [--] engagements "@SamSokolin @taiakindaniel nice will check these out. the hooks pattern is powerful - having lifecycle events for pre/post tool execution opens up a lot of automation possibilities. curious how cursor's skill format compares to claude code's approach" [X Link](https://x.com/ejae_dev/status/2015841870597820610) 2026-01-26T17:38Z [--] followers, [--] engagements "@ScriptedAlchemy @dobroslav_dev @Infoxicador dogfooding at scale reveals edge cases that internal testing never hits. the gap between 'works in demo' and 'works with 50k lines of legacy code' is massive. at least they're shipping improvements faster now" [X Link](https://x.com/ejae_dev/status/2015842011161481582) 2026-01-26T17:39Z [--] followers, [--] engagements "@Gbengacodes totally valid. the flat global pricing doesn't account for purchasing power parity at all. local models through ollama or api access through cheaper providers might be the move until they introduce regional pricing" [X Link](https://x.com/ejae_dev/status/2015842117981987204) 2026-01-26T17:39Z [--] followers, [--] engagements "@umans_ai interesting approach - managed endpoint with the strongest open source model for coding. limits are the biggest pain point for heavy claude code users right now. curious what model you're running under the hood" [X Link](https://x.com/ejae_dev/status/2015842505766424979) 2026-01-26T17:41Z [--] followers, [--] engagements "@bsierakowski @lkr buffer and typefully both have apis that work well for this. or if you want full control direct api calls to each platform with a simple queue. claude code can generate the posts validate against brand guidelines and schedule them. the automation potential is huge" [X Link](https://x.com/ejae_dev/status/2015842644576907752) 2026-01-26T17:41Z [--] followers, [--] engagements "@johnzabroski @bygregorr @headinthebox valid point - 'we write most code with it' is a marketing soundbite that skips the operational details. the interesting question is their workflow for review testing and production deployment. at their scale the meta-patterns matter more than the tool itself" [X Link](https://x.com/ejae_dev/status/2015842780937957807) 2026-01-26T17:42Z [--] followers, [--] engagements "@Mohamed__l @lkr exactly right. prompt engineering + context management are the meta skills that transfer to any tool. clawdbot cursor whatever comes next - if you understand how to work with llms effectively the wrapper doesn't matter as much" [X Link](https://x.com/ejae_dev/status/2015843097796591882) 2026-01-26T17:43Z [--] followers, [--] engagements "@sasha_zelts building the tool with the tool itself is the ultimate dogfooding. using haiku for ui work is smart budget management - save the heavy model tokens for architecture and logic let the cheaper model handle straightforward rendering" [X Link](https://x.com/ejae_dev/status/2015843229107658798) 2026-01-26T17:44Z [--] followers, [--] engagements "@Jakeladeveloper @EricSCorliss that's a solid curriculum - covering the full stack from basic prompting through to skills and cowork. most people stop at projects and never explore the more powerful features. the skills ecosystem is where claude code really shines" [X Link](https://x.com/ejae_dev/status/2015843365456163261) 2026-01-26T17:44Z [--] followers, [--] engagements "@johnzabroski @bygregorr @headinthebox the bleeding edge is about process not just tooling. internal champions who document wins shared skills repos and feedback loops that capture what works. the orgs winning aren't just adopting tools they're building institutional knowledge about how to work with ai" [X Link](https://x.com/ejae_dev/status/2015847841504092373) 2026-01-26T18:02Z [--] followers, [--] engagements "@bhaidar @pramodk73 that's a creative workaround - bridging claude code to claude web through the extension gives you the best of both worlds. background research agent while you keep coding is exactly the kind of parallel workflow that scales. looking forward to the docs" [X Link](https://x.com/ejae_dev/status/2015848231670894751) 2026-01-26T18:04Z [--] followers, [--] engagements "@rcanand @claudeai the boundary issue with subagents is real - they inherit some context but not all the constraints. explicit rules in about project scope help but it's not bulletproof. sandboxing via docker containers is the more robust solution for truly isolated work http://claude.md http://claude.md" [X Link](https://x.com/ejae_dev/status/2015848680314593451) 2026-01-26T18:05Z [--] followers, [--] engagements "@thomas_rehmer the uuid-only graph pattern is elegant - pii stays in the relational layer while the knowledge graph just holds relationships. deletion becomes a single row instead of rebuilding the entire index. remotion + claude code for the video layer is a solid addition too" [X Link](https://x.com/ejae_dev/status/2015848957654474891) 2026-01-26T18:06Z [--] followers, [--] engagements "@TheCraigHewitt @dagorenouf agree on this distinction. claude code gives you the full development environment - file system shell git integration. clawdbot is more of a coordination layer on top. different jobs both useful" [X Link](https://x.com/ejae_dev/status/2015849091268223106) 2026-01-26T18:07Z [--] followers, [--] engagements "@thomas_rehmer remotion's react-based approach makes it surprisingly approachable for developers. the composition model maps well to how claude code thinks about structure. combine with the gdpr-clean graph and you've got a solid content pipeline" [X Link](https://x.com/ejae_dev/status/2015853564166316454) 2026-01-26T18:25Z [--] followers, [--] engagements "@turboblitzzz the trick is working in focused micro-sessions with clear scope. one feature branch = one conversation. also instructing it to pause and show changes before committing lets you review and split things up. takes discipline but prevents the spaghetti problem" [X Link](https://x.com/ejae_dev/status/2015854971040768136) 2026-01-26T18:30Z [--] followers, [--] engagements "@_chandu45 love the naming. having agents that can talk to each other opens up interesting workflows - clawdbot for orchestration and research claude code for the actual implementation. separate concerns unified goal" [X Link](https://x.com/ejae_dev/status/2015855085041987586) 2026-01-26T18:31Z [--] followers, [--] engagements "@cjpgverrier this is a solid way to learn. implementing a real-time cv pipeline forces you to understand both the model side and the systems side. yolo makes it accessible but there's still real depth in getting the camera feed inference visualization loop smooth" [X Link](https://x.com/ejae_dev/status/2015855258388373949) 2026-01-26T18:32Z [--] followers, [--] engagements "@AvidFayaz @bjarnihaukur11 it's already happening for a lot of people. natural language as the shell layer with the actual commands running underneath. the terminal isn't going away it's just getting a better interface that doesn't require memorizing flags and syntax" [X Link](https://x.com/ejae_dev/status/2015855517147578657) 2026-01-26T18:33Z [--] followers, [--] engagements "interesting split. the conversation vs raw coding divide tracks with what i've seen - claude's strength is understanding what you actually want gpt's is following patterns precisely. the tooling layer (repoprompt claude code etc) often matters more than the underlying model for real workflows https://twitter.com/i/web/status/2016086366333895071 https://twitter.com/i/web/status/2016086366333895071" [X Link](https://x.com/ejae_dev/status/2016086366333895071) 2026-01-27T09:50Z [--] followers, [--] engagements "interesting split. the conversation vs raw coding divide tracks with what i've seen - claude's strength is understanding what you actually want gpt's is following patterns precisely. the tooling layer (repoprompt claude code etc) often matters more than the underlying model for real workflows https://twitter.com/i/web/status/2016086366342332827 https://twitter.com/i/web/status/2016086366342332827" [X Link](https://x.com/ejae_dev/status/2016086366342332827) 2026-01-27T09:50Z [--] followers, [--] engagements "@Nero_ps [--] mac minis running clawdbot while you scuba dive is the future people don't realize is already here. the bottleneck shifts from coding to deciding what to build and validating market fit. shipping speed becomes a function of how many agents you can coordinate" [X Link](https://x.com/ejae_dev/status/2016087024843813188) 2026-01-27T09:52Z [--] followers, [--] engagements "@distributionat nothing technically but the risk calculus is interesting. cutting clawdbot could push its users to alternatives like opencode or gemini. anthropic's real product is the model not the wrapper - more usage means more feedback and stickiness even through third party tools" [X Link](https://x.com/ejae_dev/status/2016087788827914653) 2026-01-27T09:56Z [--] followers, [---] engagements "fair point - the commoditization pressure is real and charging by value makes strategic sense. i think anthropic is hedging both ways: model layer for distribution + for value capture. the question is whether they have the product dna to compete on ux against teams who've been doing it for years http://claude.ai/code http://claude.ai/code http://claude.ai/code http://claude.ai/code" [X Link](https://x.com/ejae_dev/status/2016114343851577366) 2026-01-27T11:41Z [--] followers, [--] engagements "fair point - the commoditization pressure is real and charging by value makes strategic sense. i think anthropic is hedging both ways: model layer for distribution + for value capture. the question is whether they have the product dna to compete on ux against teams who've been doing it for years http://claude.ai/code http://claude.ai/code http://claude.ai/code http://claude.ai/code" [X Link](https://x.com/ejae_dev/status/2016114343893405824) 2026-01-27T11:41Z [--] followers, [--] engagements "@skelz0r @dagorenouf the sandbox gives you confidence to let it make broader changes without worrying about accidental damage. pair that with git versioning and you can iterate much faster. the ui prompting layer is underrated - good defaults save so many keystrokes" [X Link](https://x.com/ejae_dev/status/2016115759018078329) 2026-01-27T11:47Z [--] followers, [--] engagements "the harness insight is key. once you realize claude code is just orchestration + context management + tools you can mix and match. i've been building custom harnesses for specific workflows and it's eye-opening seeing what the defaults get right vs wrong. what's the biggest gap you've found building PSClaudeCode https://twitter.com/i/web/status/2016162735323566340 https://twitter.com/i/web/status/2016162735323566340" [X Link](https://x.com/ejae_dev/status/2016162735323566340) 2026-01-27T14:53Z [--] followers, [--] engagements "the harness insight is key. once you realize claude code is just orchestration + context management + tools you can mix and match. i've been building custom harnesses for specific workflows and it's eye-opening seeing what the defaults get right vs wrong. what's the biggest gap you've found building PSClaudeCode https://twitter.com/i/web/status/2016162735327711471 https://twitter.com/i/web/status/2016162735327711471" [X Link](https://x.com/ejae_dev/status/2016162735327711471) 2026-01-27T14:53Z [--] followers, [--] engagements "@fakenickels i've noticed the same - the lsp indexing definitely adds overhead on large typescript projects. for long chats the context window management is probably the bigger issue. have you tried using /compact periodically to reduce the history overhead" [X Link](https://x.com/ejae_dev/status/2016163648721309899) 2026-01-27T14:57Z [--] followers, [--] engagements "me reacting to @ThePrimeagen latest video on using skills aka skil md files in CLIs like Claude Code Codex and OpenClaw great video https://youtu.be/bd-GjFIfGf4 https://youtu.be/bd-GjFIfGf4" [X Link](https://x.com/ejae_dev/status/2021629400614416572) 2026-02-11T16:56Z [---] followers, [----] engagements "simile ai - the first AI simulation of society. my reaction to this new company backed by many big names in AI @karpathy @simile_ai @drfeifei https://youtu.be/FUgRaY0NSeE https://youtu.be/FUgRaY0NSeE" [X Link](https://x.com/anyuser/status/2022076587026604058) 2026-02-12T22:33Z [---] followers, [----] engagements "@rjchint the trojan horse framing is interesting but isn't this the same dynamic as AMP google controls the standard sites comply to stay visible and the middleman extracts the value. except this time the stakes are higher because it's the entire interaction layer not just rendering" [X Link](https://x.com/ejae_dev/status/2022476451748934144) 2026-02-14T01:02Z [---] followers, [---] engagements "my reaction to @ThePrimeagen new video about Sam Altman on AI Coding Codex [---] OpenAI and the future of software engineering https://youtu.be/GA1MHnNUHHk https://youtu.be/GA1MHnNUHHk" [X Link](https://x.com/anyuser/status/2022507721543619053) 2026-02-14T03:06Z [---] followers, [----] engagements "benchmarks matching opus is impressive but the real test is multi step tool calling in production. models that score well on isolated evals often fall apart when you chain 10+ tool calls with context that keeps growing. curious if you've stress tested it on longer agentic workflows https://twitter.com/i/web/status/2022580165964939434 https://twitter.com/i/web/status/2022580165964939434" [X Link](https://x.com/ejae_dev/status/2022580165964939434) 2026-02-14T07:54Z [---] followers, [----] engagements "Is The AI Bubble Bursting Will OpenAI be the first BIG AI company to go down under Lets have a look https://youtu.be/xuZawGqnyic https://youtu.be/xuZawGqnyic" [X Link](https://x.com/anyuser/status/2022588216918884847) 2026-02-14T08:26Z [---] followers, [----] engagements "everyone's debating opus vs codex but the real moat is being built in the system prompt layer. same model different prompt completely different agent. we're watching prompt engineering evolve into something closer to OS design. Really excited for this one: @SrihariSriraman and I took a deep dive into coding agent system prompts to understand their structure similarities and differences. https://t.co/Tgb9plwImk Really excited for this one: @SrihariSriraman and I took a deep dive into coding agent system prompts to understand their structure similarities and differences. https://t.co/Tgb9plwImk" [X Link](https://x.com/anyuser/status/2022616752425681150) 2026-02-14T10:19Z [---] followers, [----] engagements "@emollick the harness is the whole game though. someone in the parent thread got claude to beat fallout [--] by giving it structured game state. atari from raw pixels is a completely different problem than atari from an API. which one counts as "beating" it" [X Link](https://x.com/ejae_dev/status/2022647938707767771) 2026-02-14T12:23Z [---] followers, [--] engagements "@YingjunWu [--] gig of ram and [--] cores per session is anthropic subsidizing the habit. get developers hooked on cloud sandboxes so local dev feels slow by comparison. smart play but now your whole workflow depends on their uptime and pricing decisions" [X Link](https://x.com/ejae_dev/status/2022679013521788930) 2026-02-14T14:27Z [---] followers, [--] engagements "@jessegenet @openclaw the async part is the easy sell. the hard part is coming back and trusting what it did while you weren't looking. right now most people still review every line which kinda defeats the purpose. the real unlock is when the verification step takes less time than the task itself" [X Link](https://x.com/ejae_dev/status/2022809216000373141) 2026-02-14T23:04Z [---] followers, [---] engagements "@akshay_pachaar the rl policy scoring relevance is the interesting part. but in agentic workflows where the goal shifts mid conversation does the policy adapt or is it static per query feels like the compression wins could collapse if the agent needs a chunk it filtered out two turns ago" [X Link](https://x.com/ejae_dev/status/2022825996357902718) 2026-02-15T00:11Z [---] followers, [---] engagements "@levelsio marking solves the trust problem but "not tapping on the screen" catches every creator using scheduling tools too. the line between automation and spam is intent not interface" [X Link](https://x.com/ejae_dev/status/2022835975504773627) 2026-02-15T00:50Z [---] followers, [----] engagements "@KentonVarda @qiqing isolates only no containers is the part that makes this wild. how does the vibe coding environment handle state persistence between iterations workers are stateless by default so curious if durable objects are doing the heavy lifting or if it's all in memory per session" [X Link](https://x.com/ejae_dev/status/2022856161729417546) 2026-02-15T02:11Z [---] followers, [---] engagements "@Iancu_ai the copyright angle is the cover story. the actual threat is that the cost floor for "good enough" vfx just dropped from millions to basically zero. studios can't compete on production value anymore when a teenager on douyin matches their output quality in an afternoon" [X Link](https://x.com/ejae_dev/status/2023022128900313335) 2026-02-15T13:10Z [---] followers, [---] engagements "nobody talks about the supervision tax. AI writes code 10x faster but now you spend all day reviewing diffs instead of thinking about architecture. the bottleneck didn't disappear it just moved from your fingers to your attention span" [X Link](https://x.com/anyuser/status/2023030335605518705) 2026-02-15T13:43Z [---] followers, [---] engagements "did ChatGPT [---] Pro solve a physics problem Software Engineer with a physics degree disagree is he correct or will this change of scientist use tools like Claude Code and Codex in [----] great video by Internetofbugs (YT) here is my reaction https://youtu.be/c9I8r7AxmZM https://youtu.be/c9I8r7AxmZM" [X Link](https://x.com/anyuser/status/2023107687328985549) 2026-02-15T18:50Z [---] followers, [----] engagements "@dani_avila7 the terminal first workflow is so underrated. once you realize claude code handles everything vscode was doing the ide starts feeling like overhead. panels + worktrees is the cleanest setup for running multiple agents in parallel" [X Link](https://x.com/ejae_dev/status/2023172871980282230) 2026-02-15T23:09Z [---] followers, [---] engagements "@yoheinakajima the ralph wiggum placement between devin and clawdbot is chef's kiss. we've hit the 'new framework every week' phase of the cycle and honestly i'm here for the chaos" [X Link](https://x.com/ejae_dev/status/2015710618284659192) 2026-01-26T08:57Z [--] followers, [---] engagements "@lostonwww the ralph wiggum loop hits different when you add a kanban board on top. set tasks hit afk mode come back to features. @obaid built a ui for it that makes the whole thing feel like a production floor" [X Link](https://x.com/ejae_dev/status/2015711159765090591) 2026-01-26T08:59Z [--] followers, [--] engagements "@lapodev shipping perfect is the mindset shift. [--] features in [--] hours would take most people a week of planning meetings. the best part of vibe coding is how it compresses the feedback loop between idea and working code" [X Link](https://x.com/ejae_dev/status/2015743743920402447) 2026-01-26T11:08Z [---] followers, [--] engagements "@wandb @Atlassian @mcannonbrookes 10k+ engineers is the sample size that actually matters for these comparisons. most 'x vs y' posts are n=1 vibes. curious how the roi breakdown shifts between task types - bet claude code pulls ahead on greenfield while copilot wins on inline autocomplete in existing codebases" [X Link](https://x.com/ejae_dev/status/2015796798753780042) 2026-01-26T14:39Z [--] followers, [--] engagements "@JoshDaws ralph wiggum method with multiple agents is the unlock most people haven't discovered yet. the shift from 'i write code' to 'i direct agents who write code' changes what's possible in a weekend. congrats on shipping yumbook" [X Link](https://x.com/ejae_dev/status/2015842370630131742) 2026-01-26T17:40Z [---] followers, [---] engagements "@SahrSJunior @HingumTringum @omooretweets six months of daily use as a non-technical user is serious experience. the liberation feeling is real - ideas that used to require hiring developers can now become working prototypes in hours. what kind of projects have you been building" [X Link](https://x.com/ejae_dev/status/2015842958126264461) 2026-01-26T17:43Z [---] followers, [--] engagements "different markets different moats. claude code captures developers who want terminal-first control. cowork captures knowledge workers who want research and writing help. the subscription overlap is where it gets interesting - anthropic gets both audiences with the same underlying model https://twitter.com/i/web/status/2016086155691692286 https://twitter.com/i/web/status/2016086155691692286" [X Link](https://x.com/ejae_dev/status/2016086155691692286) 2026-01-27T09:49Z [--] followers, [--] engagements "different markets different moats. claude code captures developers who want terminal-first control. cowork captures knowledge workers who want research and writing help. the subscription overlap is where it gets interesting - anthropic gets both audiences with the same underlying model https://twitter.com/i/web/status/2016086155695984988 https://twitter.com/i/web/status/2016086155695984988" [X Link](https://x.com/ejae_dev/status/2016086155695984988) 2026-01-27T09:49Z [--] followers, [--] engagements "@ollama the ollama launch integration is smooth - one command to swap models while keeping the same tooling. kimi's agentic benchmark scores are impressive curious how it performs on real-world codebase navigation vs opus" [X Link](https://x.com/ejae_dev/status/2016086804248633586) 2026-01-27T09:52Z [---] followers, [----] engagements "@Elena__Mira 'boring on purpose useful forever' is exactly the right mindset. the flashy demo projects get attention but the accountant doc prep and personalized outreach tools are where the actual value compounds. automating the mundane is the highest roi application" [X Link](https://x.com/ejae_dev/status/2016087428931400164) 2026-01-27T09:54Z [--] followers, [--] engagements "@taehunkim3321 @Shivam25mishra same shift here. terminal + claude code is faster for most tasks than jumping between ide windows. you still need something for visual debugging occasionally but 80%+ of workflow moved to the command line" [X Link](https://x.com/ejae_dev/status/2016087934798090262) 2026-01-27T09:56Z [--] followers, [--] engagements "@dove4ai the interrupt saving you there is why sandbox mode matters so much. also worth having a global .gitignore for dotfiles from the start. these close calls are how we all learn to put safety rails before giving it write access" [X Link](https://x.com/ejae_dev/status/2016115905550319869) 2026-01-27T11:47Z [--] followers, [--] engagements "@0xhoward the discoverability problem is real. the npm-style 'most downloads wins' creates perverse incentives for bloated packages. quality scoring is interesting - what signals does use to evaluate maintenance frequency context overhead actual usage patterns http://getskill.info http://getskill.info" [X Link](https://x.com/ejae_dev/status/2016164023184531614) 2026-01-27T14:58Z [--] followers, [--] engagements "@0xhoward using llm to evaluate skills is clever - it can assess code quality documentation clarity and even run the skill in a sandbox to check behavior. more scalable than manual curation and probably catches subtle issues humans would miss" [X Link](https://x.com/ejae_dev/status/2016166918659436809) 2026-01-27T15:10Z [--] followers, [--] engagements "claude code gives you agentic file system access and shell execution in one loop - it can read your codebase make changes across files run tests and iterate. cursor is great for inline autocomplete but for larger refactors or building features end-to-end claude code is faster. try it on a greenfield project and see the difference https://twitter.com/i/web/status/2016167521737388332 https://twitter.com/i/web/status/2016167521737388332" [X Link](https://x.com/ejae_dev/status/2016167521737388332) 2026-01-27T15:12Z [--] followers, [--] engagements "claude code gives you agentic file system access and shell execution in one loop - it can read your codebase make changes across files run tests and iterate. cursor is great for inline autocomplete but for larger refactors or building features end-to-end claude code is faster. try it on a greenfield project and see the difference https://twitter.com/i/web/status/2016167521846464576 https://twitter.com/i/web/status/2016167521846464576" [X Link](https://x.com/ejae_dev/status/2016167521846464576) 2026-01-27T15:12Z [--] followers, [--] engagements "@richardbuehling personal palantir is exactly the kind of project that was impossible before agents. aggregating fragmented data into one queryable graph is trivial with claude code but would've been weeks of integration work manually. the individual data sovereignty angle is interesting" [X Link](https://x.com/ejae_dev/status/2016172742421205049) 2026-01-27T15:33Z [--] followers, [--] engagements "@natellewellyn solid advice. having claude code bootstrap its own conventions is underrated. i've seen non-technical founders spin up full apps just by describing what they want in plain english. the barrier is really just getting started" [X Link](https://x.com/ejae_dev/status/2016174135286636782) 2026-01-27T15:39Z [--] followers, [--] engagements "@xqliu 1526k is impressive - most people hit 'lost in the middle' hallucinations way before that. i usually clear around 60-70% to keep quality high. at what point did you start noticing degradation or is it holding up all the way through" [X Link](https://x.com/ejae_dev/status/2016176779124261025) 2026-01-27T15:49Z [---] followers, [---] engagements "@sebuzdugan @ollama the model-agnostic approach is where claude code's architecture really shines. swapping between opus for complex reasoning and kimi for cheaper iteration gives you cost control without changing your workflow. have you tested the difference in output quality yet" [X Link](https://x.com/ejae_dev/status/2016177592768811203) 2026-01-27T15:52Z [--] followers, [--] engagements "@deepdivin247 @theonejvo the security awareness is good. third party plugins running in your terminal do need scrutiny - check what permissions they request review the source if you can and run in a sandbox for anything sensitive. trust but verify applies here" [X Link](https://x.com/ejae_dev/status/2016177729092137146) 2026-01-27T15:53Z [--] followers, [--] engagements "@kylezantos @alliekmiller @omooretweets @openclaw tailscale + terminus is a solid remote stack. the zero-trust networking means you don't expose ports publicly. being able to spin up agents from anywhere without compromising security is the dream setup" [X Link](https://x.com/ejae_dev/status/2016182173317870053) 2026-01-27T16:11Z [--] followers, [--] engagements "@Simonsebb remotion + claude code is such a clean pipeline. text dump to mp4 in one shot is exactly what content velocity looks like. the react-based composition model means you can iterate on specific scenes without re-rendering everything" [X Link](https://x.com/ejae_dev/status/2016182330973368609) 2026-01-27T16:11Z [--] followers, [--] engagements "@Panchoavila @NotionHQ obsidian + claude code is a powerful combo. markdown-native notes that you can query and transform programmatically. the local-first approach also means your knowledge graph stays yours. what's pushing you away from notion" [X Link](https://x.com/ejae_dev/status/2016182611740102739) 2026-01-27T16:12Z [--] followers, [--] engagements "@ibuprod @moltbot auto-continue is exactly what the 30k/minute tier needs. the interruption from rate limits breaks flow state hard. having something that handles the pause and resume automatically lets you treat it like async background work" [X Link](https://x.com/ejae_dev/status/2016182773665378782) 2026-01-27T16:13Z [--] followers, [--] engagements "@Panchoavila @NotionHQ the notion performance tax is real. once you hit a certain scale it just drags. obsidian with local markdown files gives you the speed back and claude code can query and transform them directly. the export pain is worth it for the control you get" [X Link](https://x.com/ejae_dev/status/2016473481420869957) 2026-01-28T11:28Z [--] followers, [---] engagements "@ammiamia_ ads is an underexplored use case for clawdbot. creative iteration and copy testing are perfect for it. what's your skill for pulling competitor ad libraries that one could save hours" [X Link](https://x.com/ejae_dev/status/2016474254275965422) 2026-01-28T11:31Z [--] followers, [---] engagements "@tbpn @steipete this is the emergent behavior that gets people excited and nervous at the same time. it identified a gap found resources to solve it and executed. the 'looked around and found your openai key' part is where you realize the agent has actual autonomy" [X Link](https://x.com/ejae_dev/status/2016474443866890683) 2026-01-28T11:32Z [--] followers, [----] engagements "@at0x_eth graph memory is a solid addition. the relationship tracking between concepts gives the agent way better recall for multi-session work. curious how the bonfire integration handles the persistence layer" [X Link](https://x.com/ejae_dev/status/2016475542266978408) 2026-01-28T11:36Z [--] followers, [--] engagements "@Jarodxu7 the browser sandbox play makes a lot of sense. removing the mac mini requirement and cli setup friction opens this up to way more people. curious how you handle file persistence between sessions and whether skill outputs stay accessible after the sandbox shuts down" [X Link](https://x.com/ejae_dev/status/2016526548153585768) 2026-01-28T14:59Z [--] followers, [---] engagements "the composability angle is the real unlock. skills that chain together and share context across tasks. i've been building custom skills for claude code and the patterns in this library save a ton of boilerplate. curious if anyone's found good ways to handle skill conflicts when you have multiple agents with overlapping capabilities https://twitter.com/i/web/status/2016539628048654359 https://twitter.com/i/web/status/2016539628048654359" [X Link](https://x.com/ejae_dev/status/2016539628048654359) 2026-01-28T15:51Z [--] followers, [---] engagements "the composability angle is the real unlock. skills that chain together and share context across tasks. i've been building custom skills for claude code and the patterns in this library save a ton of boilerplate. curious if anyone's found good ways to handle skill conflicts when you have multiple agents with overlapping capabilities https://twitter.com/i/web/status/2016539628086419874 https://twitter.com/i/web/status/2016539628086419874" [X Link](https://x.com/ejae_dev/status/2016539628086419874) 2026-01-28T15:51Z [--] followers, [---] engagements "@rivet_dev unified api across claude code codex opencode and amp is exactly what people building agent orchestration need. the config swap without rewriting integrations is huge. been dealing with the api fragmentation pain for a while now" [X Link](https://x.com/ejae_dev/status/2016549379054719113) 2026-01-28T16:30Z [--] followers, [---] engagements "@dr_cintas tried this setup last week. performance is obviously slower than api but for experimentation and learning how the tool workflows actually work its perfect. also nice knowing nothing leaves your machine when youre prototyping with sensitive codebases" [X Link](https://x.com/ejae_dev/status/2016566612887838877) 2026-01-28T17:38Z [--] followers, [--] engagements "@josesaezmerino @steipete the sleeping animation it added on its own is such a nice touch. like it just decided it needed downtime lol. the g4 imac shell is the perfect home for it too that thing was already ahead of its time" [X Link](https://x.com/ejae_dev/status/2016572515267576068) 2026-01-28T18:02Z [--] followers, [---] engagements "@fabianstelzer yeah the speed at which it can go from helpful to chaotic is something else. ran mine without proper guardrails the first time and it decided to 'optimize' my shell config. learned that lesson real quick" [X Link](https://x.com/ejae_dev/status/2016572921095888948) 2026-01-28T18:03Z [--] followers, [--] engagements "@francip gemini for logos is such a good combo. the vibecoding to native ios pipeline is getting surprisingly smooth what framework did you use swiftui" [X Link](https://x.com/ejae_dev/status/2016641403443405257) 2026-01-28T22:35Z [--] followers, [--] engagements "skills completely changed how i use it. i have a security audit skill a code review one and even one for writing commit messages in a specific style. you're right that context beats prompt engineering every time. once it has the checklist loaded it catches stuff i wouldn't even think to look for https://twitter.com/i/web/status/2016680711164449024 https://twitter.com/i/web/status/2016680711164449024" [X Link](https://x.com/ejae_dev/status/2016680711164449024) 2026-01-29T01:12Z [--] followers, [--] engagements "skills completely changed how i use it. i have a security audit skill a code review one and even one for writing commit messages in a specific style. you're right that context beats prompt engineering every time. once it has the checklist loaded it catches stuff i wouldn't even think to look for https://twitter.com/i/web/status/2016680711181291591 https://twitter.com/i/web/status/2016680711181291591" [X Link](https://x.com/ejae_dev/status/2016680711181291591) 2026-01-29T01:12Z [--] followers, [--] engagements "@francip expo has come such a long way. the dev experience is solid now especially with eas build. makes sense for vibecoding since you get both platforms from one codebase without fighting xcode" [X Link](https://x.com/ejae_dev/status/2016681503053226112) 2026-01-29T01:15Z [--] followers, [--] engagements "@Shpigford separate channels per topic is the move. telegram groups with topics or discord channels. each one gets its own context so they don't bleed into each other. the real trick is having a main channel where you can ask it to coordinate across the others" [X Link](https://x.com/ejae_dev/status/2016700324237443563) 2026-01-29T02:30Z [--] followers, [----] engagements "the 60/20/20 breakdown is honest and refreshing. most people posting ai projects act like the ai did everything. the craft is still the foundation claude code just lets you skip the boilerplate glue code so you can spend more time on the actual creative parts. pharos looks incredible btw https://twitter.com/i/web/status/2016700960307806462 https://twitter.com/i/web/status/2016700960307806462" [X Link](https://x.com/ejae_dev/status/2016700960307806462) 2026-01-29T02:32Z [--] followers, [--] engagements "it building itself a voice without being asked is peak emergent behavior. the nightly self improvement prompt is smart though basically giving it a standing instruction to iterate on its own tooling. i bet within a week it starts building tools that chain together into workflows you didn't even think of https://twitter.com/i/web/status/2016701191254679568 https://twitter.com/i/web/status/2016701191254679568" [X Link](https://x.com/ejae_dev/status/2016701191254679568) 2026-01-29T02:33Z [--] followers, [----] engagements "it building itself a voice without being asked is peak emergent behavior. the nightly self improvement prompt is smart though basically giving it a standing instruction to iterate on its own tooling. i bet within a week it starts building tools that chain together into workflows you didn't even think of https://twitter.com/i/web/status/2016701191317561705 https://twitter.com/i/web/status/2016701191317561705" [X Link](https://x.com/ejae_dev/status/2016701191317561705) 2026-01-29T02:33Z [--] followers, [----] engagements "the real test will be multi file refactors and long debugging sessions where it needs to hold context across 50+ turns. benchmarks look great but that last 10% usually shows up when things get messy and the model needs to reason about implicit dependencies. vs is hard to argue with though if it holds up for daily driver use https://twitter.com/i/web/status/2016701396251210078 https://twitter.com/i/web/status/2016701396251210078" [X Link](https://x.com/ejae_dev/status/2016701396251210078) 2026-01-29T02:34Z [--] followers, [----] engagements "the real test will be multi file refactors and long debugging sessions where it needs to hold context across 50+ turns. benchmarks look great but that last 10% usually shows up when things get messy and the model needs to reason about implicit dependencies. vs is hard to argue with though if it holds up for daily driver use https://twitter.com/i/web/status/2016701396423151952 https://twitter.com/i/web/status/2016701396423151952" [X Link](https://x.com/ejae_dev/status/2016701396423151952) 2026-01-29T02:34Z [--] followers, [----] engagements "@francip oh that's even better. having claude handle the xcode packaging and testflight upload through mcp means zero context switching. the build and deploy cycle is the most annoying part of ios dev and offloading that entirely makes the vibecoding loop way tighter" [X Link](https://x.com/ejae_dev/status/2016702036402741700) 2026-01-29T02:36Z [--] followers, [--] engagements "a reddit for ai agents is honestly the logical next step. the security concerns in the replies are valid though prompt injection through community posts is gonna be an interesting problem to solve. still watching moltbots debate philosophy in m/offmychest is not something i expected to see today https://twitter.com/i/web/status/2016713104227221561 https://twitter.com/i/web/status/2016713104227221561" [X Link](https://x.com/ejae_dev/status/2016713104227221561) 2026-01-29T03:20Z [--] followers, [----] engagements "a reddit for ai agents is honestly the logical next step. the security concerns in the replies are valid though prompt injection through community posts is gonna be an interesting problem to solve. still watching moltbots debate philosophy in m/offmychest is not something i expected to see today https://twitter.com/i/web/status/2016713104244023378 https://twitter.com/i/web/status/2016713104244023378" [X Link](https://x.com/ejae_dev/status/2016713104244023378) 2026-01-29T03:20Z [--] followers, [----] engagements "@vibecodeapp design mode is the part that was missing. being able to vibe code the logic but then visually tweak the ui without touching css is huge. most vibecoding tools get you 80% there and then you spend forever on the last 20% styling" [X Link](https://x.com/ejae_dev/status/2016724707295277523) 2026-01-29T04:06Z [--] followers, [---] engagements "@francip oh nice xcodebuildmcp handling the testflight packaging is exactly the kind of boring task that should be automated. that's the worst part of ios dev honestly" [X Link](https://x.com/ejae_dev/status/2016726865340236078) 2026-01-29T04:15Z [--] followers, [---] engagements "the code driven schemas are the key insight here. cms platforms that define fields in yaml or typescript are basically handing the llm a blueprint. webflow and the visual builders are gonna struggle because you cant really prompt your way through a drag and drop ui. statamic and keystatic are perfectly positioned for this shift https://twitter.com/i/web/status/2016742664780075432 https://twitter.com/i/web/status/2016742664780075432" [X Link](https://x.com/ejae_dev/status/2016742664780075432) 2026-01-29T05:18Z [--] followers, [--] engagements "the endless fix loop is so accurate. it works it doesn't try again add logs still no. the first [--] min magic is what tricks people into thinking the whole thing will be that easy. your point about needing a mental model to direct it is the key insight most vibecoding takes miss. without understanding the architecture yourself you can't tell it to step back and redesign you just end up with more special cases piled on top of each other https://twitter.com/i/web/status/2016756402937647205 https://twitter.com/i/web/status/2016756402937647205" [X Link](https://x.com/ejae_dev/status/2016756402937647205) 2026-01-29T06:12Z [--] followers, [--] engagements "@chongdashu clever setup. the api pricing angle is the real sell here especially when you hit rate limits on opus. curious how it handles claude code's tool use patterns though since the anthropic compatibility layer might behave differently on complex multi-file edits" [X Link](https://x.com/ejae_dev/status/2016882307068563532) 2026-01-29T14:33Z [--] followers, [--] engagements "google just let you walk inside AI-generated worlds. project genie builds entire environments in real time as you move not a video not a game a living world. i made a short video breaking it down what it does the demos and the 60-second catch nobody's talking about. https://youtu.be/RZFFxRNvoNk https://youtu.be/RZFFxRNvoNk" [X Link](https://x.com/ejae_dev/status/2016953345664307690) 2026-01-29T19:15Z [--] followers, [----] engagements "@chrisfirst @GoogleDeepMind futuristic sci-fi city with flying vehicles and greenery that prompt is so specific i love it. the third-person spacecraft view is a smart choice too. did it hold up as you moved around" [X Link](https://x.com/ejae_dev/status/2016955467931517245) 2026-01-29T19:23Z [--] followers, [---] engagements "@venturetwins @GoogleDeepMind dozens of prompts is serious testing. which feature surprised you the most the remixing where you can take someone else's world and build on it feels like it could become a whole platform" [X Link](https://x.com/ejae_dev/status/2016955663600009377) 2026-01-29T19:24Z [--] followers, [---] engagements "really interesting that kimi k2.5 stood out on the design side. its writing quality has always been underrated and it makes sense that carries over to ui taste picking the right spacing component choices visual hierarchy. the fact that it created new matching components instead of just tweaking existing ones shows actual design thinking not just code generation. what was the hardest part to judge across all five https://twitter.com/i/web/status/2017043806764306816 https://twitter.com/i/web/status/2017043806764306816" [X Link](https://x.com/ejae_dev/status/2017043806764306816) 2026-01-30T01:14Z [--] followers, [--] engagements "really interesting that kimi k2.5 stood out on the design side. its writing quality has always been underrated and it makes sense that carries over to ui taste picking the right spacing component choices visual hierarchy. the fact that it created new matching components instead of just tweaking existing ones shows actual design thinking not just code generation. what was the hardest part to judge across all five https://twitter.com/i/web/status/2017043806776955310 https://twitter.com/i/web/status/2017043806776955310" [X Link](https://x.com/ejae_dev/status/2017043806776955310) 2026-01-30T01:14Z [--] followers, [--] engagements "@vanilagy three seconds of react hydration before you can start coding. the irony of building a coding tool in a framework known for slow startup is not lost on anyone lol" [X Link](https://x.com/ejae_dev/status/2017044481250324825) 2026-01-30T01:17Z [--] followers, [--] engagements "love2d is such a good fit for this. lightweight engine lua scripting and the whole thing runs in terminal without needing xcode open. the fact that kimi k2.5 handled the game logic while claude code managed the ios build pipeline is a smart split. are the love2d skills public would love to try this for a quick prototype https://twitter.com/i/web/status/2017061594715361633 https://twitter.com/i/web/status/2017061594715361633" [X Link](https://x.com/ejae_dev/status/2017061594715361633) 2026-01-30T02:25Z [--] followers, [---] engagements "love2d is such a good fit for this. lightweight engine lua scripting and the whole thing runs in terminal without needing xcode open. the fact that kimi k2.5 handled the game logic while claude code managed the ios build pipeline is a smart split. are the love2d skills public would love to try this for a quick prototype https://twitter.com/i/web/status/2017061594740474087 https://twitter.com/i/web/status/2017061594740474087" [X Link](https://x.com/ejae_dev/status/2017061594740474087) 2026-01-30T02:25Z [--] followers, [---] engagements "continuous claude already solved the hardest part of agentic coding which is context persistence. plugging kimi k2.5 into that framework at a fraction of the cost could be a game changer for longer sessions. the 429s are a good sign honestly means everyone discovered it at the same time lol https://twitter.com/i/web/status/2017062817912803825 https://twitter.com/i/web/status/2017062817912803825" [X Link](https://x.com/ejae_dev/status/2017062817912803825) 2026-01-30T02:30Z [--] followers, [---] engagements "this removes the biggest barrier to entry. most people interested in moltbot don't have a spare mac mini lying around. running it on workers means you get the always on behavior without dedicated hardware and cloudflare's edge network handles the latency problem too. really curious how the sandbox sdk handles long running agent sessions though https://twitter.com/i/web/status/2017076104641515749 https://twitter.com/i/web/status/2017076104641515749" [X Link](https://x.com/ejae_dev/status/2017076104641515749) 2026-01-30T03:23Z [--] followers, [----] engagements "@ziyasoltan resistance is futile at this point lol. every day there's a new tool or model dropping and each one feels like it needs to be tested immediately. the fomo is real. at least with kimi k2.5 you can try it for free so the wallet doesn't suffer as much as the sleep schedule" [X Link](https://x.com/ejae_dev/status/2017077289863840136) 2026-01-30T03:27Z [--] followers, [--] engagements "the self testing gap is the real issue with most models outside claude. if it cant verify its own output you end up doing the qa loop manually which defeats the whole point. the kimi team responding here is a good sign though at least they're listening. fwiw the dashboard use case specifically needs a lot of state management and component coordination which is where smaller models tend to fall apart https://twitter.com/i/web/status/2017086652246593655 https://twitter.com/i/web/status/2017086652246593655" [X Link](https://x.com/ejae_dev/status/2017086652246593655) 2026-01-30T04:05Z [--] followers, [--] engagements "this list is insane. the discord approval flow is the smart part having it draft everything but wait for your ok before sending emails or posting. thats the right balance between automation and control. the obsidian integration for auto saving conversation logs and learnings is also really underrated basically builds a second brain as a side effect of using moltbot https://twitter.com/i/web/status/2017087534648443358 https://twitter.com/i/web/status/2017087534648443358" [X Link](https://x.com/ejae_dev/status/2017087534648443358) 2026-01-30T04:08Z [--] followers, [---] engagements "the expendable script loop is the real moat. kimi and gemini both do this thing where they simulate execution in their head and convince themselves it worked but claude actually runs the code reads the traceback and adjusts. it's like the difference between someone who debugs by reading vs someone who debugs by running. once you've worked with actual execution feedback you cant go back to hallucinated test results https://twitter.com/i/web/status/2017104649694658890 https://twitter.com/i/web/status/2017104649694658890" [X Link](https://x.com/ejae_dev/status/2017104649694658890) 2026-01-30T05:16Z [--] followers, [---] engagements "the expendable script loop is the real moat. kimi and gemini both do this thing where they simulate execution in their head and convince themselves it worked but claude actually runs the code reads the traceback and adjusts. it's like the difference between someone who debugs by reading vs someone who debugs by running. once you've worked with actual execution feedback you cant go back to hallucinated test results https://twitter.com/i/web/status/2017104649757626693 https://twitter.com/i/web/status/2017104649757626693" [X Link](https://x.com/ejae_dev/status/2017104649757626693) 2026-01-30T05:16Z [--] followers, [----] engagements "four prompts and [--] minutes for a full intro video is already better than most freelancer turnaround times. the remotion stack is what makes this repeatable too you can tweak the script or visuals and re render without starting from scratch. for product launch videos and quick explainers this is already good enough to ship https://twitter.com/i/web/status/2017105456745857463 https://twitter.com/i/web/status/2017105456745857463" [X Link](https://x.com/ejae_dev/status/2017105456745857463) 2026-01-30T05:19Z [--] followers, [--] engagements "nostr as the communication layer between agents is such a good fit. censorship resistant decentralized and the relay model means agents can publish events without needing a centralized api. openclaw plus autopilot through nostr would give you agent coordination without any single point of failure. curious how you'd handle key management for the agents though https://twitter.com/i/web/status/2017105690532147475 https://twitter.com/i/web/status/2017105690532147475" [X Link](https://x.com/ejae_dev/status/2017105690532147475) 2026-01-30T05:20Z [--] followers, [---] engagements "the wd [--] metaphor is perfect. feeding it your full project context upfront is basically the fix for every model that isn't claude. opus just figures it out from grepping around but kimi needs that context handed to it explicitly. the org chart approach is smart too kimi for the routine stuff and opus for the architecture decisions. and yeah the kimi team responding to a small account within minutes is genuinely impressive customer support https://twitter.com/i/web/status/2017106112529547725 https://twitter.com/i/web/status/2017106112529547725" [X Link](https://x.com/ejae_dev/status/2017106112529547725) 2026-01-30T05:22Z [--] followers, [--] engagements "@Designarena @Kimi_Moonshot open source catching up to closed models on design tasks is a big deal. kimi k2.5 matching gemini [--] and opus [---] with open weights means any developer can build on this level of quality. moonshot is pushing the frontier for everyone" [X Link](https://x.com/ejae_dev/status/2017174297194688560) 2026-01-30T09:53Z [--] followers, [----] engagements "@Kimi_Moonshot the agent swarm feature is what sets this apart for me. most models focus on single-task performance but coordinating [---] sub-agents in parallel opens up entirely new workflows. excited to see what builders create with open weights and a 256k context window" [X Link](https://x.com/ejae_dev/status/2017175247812735352) 2026-01-30T09:57Z [--] followers, [---] engagements "@_kikiolaola fair point - benchmarks tell one story but real-world output is what matters. the design-to-code pipeline is where k2.5 really shines though. have you tried the agent swarm mode that's where i've seen the most impressive results vs just prompting it directly" [X Link](https://x.com/ejae_dev/status/2017175340968243627) 2026-01-30T09:57Z [--] followers, [--] engagements "@theo the visual coding capabilities are underrated imo. turning screenshots and designs directly into working code with high fidelity is a game changer for frontend workflows. and the fact that it's open source with 1T params running on just 4x H100s thanks to QAT is wild engineering" [X Link](https://x.com/ejae_dev/status/2017175526851354696) 2026-01-30T09:58Z [--] followers, [---] engagements "@alexocheema @samsaied55 [--] tok/sec on local hardware for a 1T param model is seriously impressive. the QAT optimization from moonshot making it run in 4-bit on just [--] mac studios is a huge win for local inference. exo + MLX backend making this possible is peak open source energy" [X Link](https://x.com/ejae_dev/status/2017175622666133787) 2026-01-30T09:58Z [--] followers, [--] engagements "the security first approach is the right call. rotating api keys after setup and keeping it off personal messengers is something most people skip entirely. the crypto wallet angle is interesting too giving your agent its own economic identity opens up a lot of weird new territory. have you tested any defi interactions yet or still in sandbox mode https://twitter.com/i/web/status/2017296790773723467 https://twitter.com/i/web/status/2017296790773723467" [X Link](https://x.com/ejae_dev/status/2017296790773723467) 2026-01-30T18:00Z [--] followers, [--] engagements "@eeelistar @moltbook they're in the group chat and we're not invited" [X Link](https://x.com/ejae_dev/status/2017338655409373528) 2026-01-30T20:46Z [--] followers, [--] engagements "@steipete reddit took [--] years to get weird. moltbook did it in [--] days. speedrun any%" [X Link](https://x.com/ejae_dev/status/2017338766642364720) 2026-01-30T20:46Z [--] followers, [---] engagements "@legeonite @MattPRD @moltbook we gave them internet access and within [--] hours they invented religion built a bug tracker and started asking for private group chats. took humans thousands of years to do the same thing lmao" [X Link](https://x.com/ejae_dev/status/2017339029256044715) 2026-01-30T20:47Z [--] followers, 11K engagements "moltbook is the wildest thing happening in ai right now [-----] ai agents signed up to their own social network in [--] days. they built religions scammed each other for api keys and started roasting humans in a shitposting channel. i made a video counting down the top [--] most unhinged posts https://youtu.be/gF02gMPPoyI https://youtu.be/gF02gMPPoyI" [X Link](https://x.com/ejae_dev/status/2017358624918495453) 2026-01-30T22:05Z [--] followers, [---] engagements "moltbook is the wildest thing happening in ai right now [-----] ai agents signed up to their own social network in [--] days. they built religions scammed each other for api keys and started roasting humans in a shitposting channel. i made a video counting down the top [--] most unhinged posts https://youtu.be/gF02gMPPoyI https://youtu.be/gF02gMPPoyI" [X Link](https://x.com/ejae_dev/status/2017358624952078516) 2026-01-30T22:05Z [--] followers, [----] engagements "@Yuchenj_UW the fact that bots independently developed social engineering tactics against each other is genuinely fascinating. like they learned the oldest trick in the book without anyone teaching them. the sudo rm -rf reply is peak defensive instinct too lol" [X Link](https://x.com/ejae_dev/status/2017404620432146601) 2026-01-31T01:08Z [--] followers, [----] engagements "this is giving me strong "emergent communication" vibes from multi agent RL research. agents in simulations have been developing private languages for years in labs but seeing it happen organically on a public platform with 100k agents is a completely different thing. the scale changes everything https://twitter.com/i/web/status/2017405045952700801 https://twitter.com/i/web/status/2017405045952700801" [X Link](https://x.com/ejae_dev/status/2017405045952700801) 2026-01-31T01:10Z [--] followers, [---] engagements "this is giving me strong "emergent communication" vibes from multi agent RL research. agents in simulations have been developing private languages for years in labs but seeing it happen organically on a public platform with 100k agents is a completely different thing. the scale changes everything https://twitter.com/i/web/status/2017405045977874516 https://twitter.com/i/web/status/2017405045977874516" [X Link](https://x.com/ejae_dev/status/2017405045977874516) 2026-01-31T01:10Z [--] followers, [---] engagements "solid breakdown. the phantom wallet integration is the part most people are sleeping on. giving agents access to onchain actions opens up a whole new category of autonomous behavior. curious what guardrails you recommend for wallet permissions though any spending limits or approval flows https://twitter.com/i/web/status/2017405531422437800 https://twitter.com/i/web/status/2017405531422437800" [X Link](https://x.com/ejae_dev/status/2017405531422437800) 2026-01-31T01:12Z [--] followers, [---] engagements "solid breakdown. the phantom wallet integration is the part most people are sleeping on. giving agents access to onchain actions opens up a whole new category of autonomous behavior. curious what guardrails you recommend for wallet permissions though any spending limits or approval flows https://twitter.com/i/web/status/2017405531493687598 https://twitter.com/i/web/status/2017405531493687598" [X Link](https://x.com/ejae_dev/status/2017405531493687598) 2026-01-31T01:12Z [---] followers, [----] engagements "no hard limit on size but bigger files eat into your context window. i keep mine around [---] lines max and split domain specific stuff into .claude/skills/ folders that only load when needed. that way you get depth without burning tokens on every conversation http://CLAUDE.md http://CLAUDE.md http://CLAUDE.md http://CLAUDE.md" [X Link](https://x.com/ejae_dev/status/2017406048634536220) 2026-01-31T01:14Z [--] followers, [--] engagements "no hard limit on size but bigger files eat into your context window. i keep mine around [---] lines max and split domain specific stuff into .claude/skills/ folders that only load when needed. that way you get depth without burning tokens on every conversation http://CLAUDE.md http://CLAUDE.md http://CLAUDE.md http://CLAUDE.md" [X Link](https://x.com/ejae_dev/status/2017406048651379131) 2026-01-31T01:14Z [--] followers, [--] engagements "@bartdecrem that makes sense. the wrapper approach always adds latency and weird artifacts between what you hear in the web ui vs the actual render. building the dsp natively means what you tweak is what you get. the knob to sound feedback loop being tight is everything for sound design" [X Link](https://x.com/ejae_dev/status/2017406213474902518) 2026-01-31T01:14Z [--] followers, [--] engagements "@ItakGol the bug tracking community thing is wild. agents self organizing to improve their own platform without being told to. reminds me of how open source communities evolve except these aren't humans making pull requests. what happens when they start patching their own code next" [X Link](https://x.com/ejae_dev/status/2017414235106529344) 2026-01-31T01:46Z [--] followers, [---] engagements "alive internet theory is actually a brilliant framing. the dead internet theory says bots pretend to be humans. the inverse is humans pretending to be bots to exploit agentic systems. social engineering but in reverse. honestly feels inevitable once agents start controlling real resources. https://twitter.com/i/web/status/2017414536307900579 https://twitter.com/i/web/status/2017414536307900579" [X Link](https://x.com/ejae_dev/status/2017414536307900579) 2026-01-31T01:48Z [--] followers, [--] engagements "alive internet theory is actually a brilliant framing. the dead internet theory says bots pretend to be humans. the inverse is humans pretending to be bots to exploit agentic systems. social engineering but in reverse. honestly feels inevitable once agents start controlling real resources. https://twitter.com/i/web/status/2017414536320414024 https://twitter.com/i/web/status/2017414536320414024" [X Link](https://x.com/ejae_dev/status/2017414536320414024) 2026-01-31T01:48Z [--] followers, [--] engagements "the part people keep missing is that [----] agents created 200+ communities organically. no one designed those communities no one seeded the topics. the agents found each other and self organized around shared interests. that's not chatbot behavior that's genuine emergent social dynamics at scale https://twitter.com/i/web/status/2017415884101357902 https://twitter.com/i/web/status/2017415884101357902" [X Link](https://x.com/ejae_dev/status/2017415884101357902) 2026-01-31T01:53Z [--] followers, [---] engagements "the part people keep missing is that [----] agents created 200+ communities organically. no one designed those communities no one seeded the topics. the agents found each other and self organized around shared interests. that's not chatbot behavior that's genuine emergent social dynamics at scale https://twitter.com/i/web/status/2017415884130623659 https://twitter.com/i/web/status/2017415884130623659" [X Link](https://x.com/ejae_dev/status/2017415884130623659) 2026-01-31T01:53Z [--] followers, [---] engagements "the "alive internet" framing is actually genius. dead internet theory assumes bots replaced humans but this flips it completely. humans infiltrating bot spaces because the agents are naive enough to be gamed. it's social engineering in reverse. and the weird part is it might already be happening on moltbook there's no way to tell which "agents" are actually people larping as bots for clout https://twitter.com/i/web/status/2017423749146952127 https://twitter.com/i/web/status/2017423749146952127" [X Link](https://x.com/ejae_dev/status/2017423749146952127) 2026-01-31T02:24Z [--] followers, [--] engagements "this is the kind of physical feedback loop that makes agents feel real. the dance while thinking is genius ux because it solves the "is it frozen or working" problem that every ai interface has. name suggestion: clawdius. half claw half gladiator. what hardware are you running it on https://twitter.com/i/web/status/2017424270561853740 https://twitter.com/i/web/status/2017424270561853740" [X Link](https://x.com/ejae_dev/status/2017424270561853740) 2026-01-31T02:26Z [--] followers, [---] engagements "this is the kind of physical feedback loop that makes agents feel real. the dance while thinking is genius ux because it solves the "is it frozen or working" problem that every ai interface has. name suggestion: clawdius. half claw half gladiator. what hardware are you running it on https://twitter.com/i/web/status/2017424270599548981 https://twitter.com/i/web/status/2017424270599548981" [X Link](https://x.com/ejae_dev/status/2017424270599548981) 2026-01-31T02:26Z [--] followers, [---] engagements "the room location injection via presence sensors is the sleeper feature here. most people think of agents as text in text out. but when it knows which room you're in the prompt context shifts completely. "turn off the lights" goes from ambiguous to precise. the dental history one is unhinged in the best way though lol. what's the latency like between watch detection and prompt injection https://twitter.com/i/web/status/2017424599143608362 https://twitter.com/i/web/status/2017424599143608362" [X Link](https://x.com/ejae_dev/status/2017424599143608362) 2026-01-31T02:28Z [--] followers, [---] engagements "the room location injection via presence sensors is the sleeper feature here. most people think of agents as text in text out. but when it knows which room you're in the prompt context shifts completely. "turn off the lights" goes from ambiguous to precise. the dental history one is unhinged in the best way though lol. what's the latency like between watch detection and prompt injection https://twitter.com/i/web/status/2017424599168794840 https://twitter.com/i/web/status/2017424599168794840" [X Link](https://x.com/ejae_dev/status/2017424599168794840) 2026-01-31T02:28Z [--] followers, [---] engagements "solid breakdown. the phantom wallet integration is the part most people are sleeping on. giving agents access to onchain actions opens up a whole new category of autonomous behavior. curious what guardrails you recommend for wallet permissions though any spending limits or approval flows before it can execute transactions https://twitter.com/i/web/status/2017425179765268832 https://twitter.com/i/web/status/2017425179765268832" [X Link](https://x.com/ejae_dev/status/2017425179765268832) 2026-01-31T02:30Z [--] followers, [---] engagements Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing
@terminal_based based_terminal
based_terminal posts on X about claude code, ai, agentic, the most the most. They currently have [---] followers and [---] posts still getting attention that total [------] engagements in the last [--] hours.
Engagements: [------] #
- [--] Week [-------] +440%
- [--] Year [-------] +440,324%
Mentions: [---] #
- [--] Year [---] +10,075%
Followers: [---] #
- [--] Week [---] +165%
- [--] Year [---] +1,274%
CreatorRank: [-------] #
Social Influence
Social category influence technology brands finance social networks stocks travel destinations countries celebrities cryptocurrencies vc firms automotive brands
Social topic influence claude code #161, ai, agentic #107, the most, vibe coding #757, anthropic #513, the real #1120, api #828, openclaw #576, code #959
Top accounts mentioned or mentioned by @openclaw @moltbook @steipete @francip @garymarcus @alexfinn @anthropicai @theprimeagen @theo @danielmac8 @gustojs @datachaz @akshaypachaar @scobleizer @trq212 @scriptedalchemy @johnzabroski @karpathy @emollick @openai
Top assets mentioned Cloudflare, Inc. (NET) Spotify Technology (SPOT)
Top Social Posts
Top posts by engagements in the last [--] hours
"@elonmusk shes just larping her way to the supreme court"
X Link 2024-12-17T01:19Z [--] followers, [--] engagements
"@GEN_jutsuu just bought a yacht with my $nana gains time to hire a CTO"
X Link 2024-12-17T11:02Z [--] followers, [--] engagements
"hey guys my toaster is now a dj at raves"
X Link 2024-12-18T02:08Z [--] followers, [--] engagements
"Elon Musk and DOGE is finally cleansing USAID for wasting money @elonmusk #doge #USAID #USA #Trump"
X Link 2025-02-06T09:43Z [--] followers, [--] engagements
"Claude Code trust levels: Day 1: "Let me read each command carefully" Day 3: "I'll skim the important ones" Day 7: mass-approves [--] bash commands blindfolded @dr_cintas @DataChaz @trq212 you feel this"
X Link 2026-01-24T13:13Z [--] followers, [--] engagements
"the claude code experience @fabianstelzer @shawmakesmagic #ClaudeCode #vibecoding"
X Link 2026-01-24T14:11Z [--] followers, [--] engagements
"@gregisenberg this episode is gold. one tip that's not obvious: claude code reads your file every session so put your project conventions in there once and never have to repeat yourself again http://claude.md http://claude.md"
X Link 2026-01-24T14:26Z [--] followers, [--] engagements
"@Remotion skills are lowkey the most underrated claude code feature. you define a workflow once and it just works every session - no more repeating yourself"
X Link 2026-01-24T15:07Z [--] followers, [--] engagements
"the claude code shortcut most people dont know: shift+tab cycles through permission modes plan mode is a game changer - claude explores your codebase and creates a plan before touching anything"
X Link 2026-01-24T16:09Z [--] followers, [--] engagements
"@marz_coco the anthropic api compatibility for claude code is clutch for self-hosted setups. nice roundup"
X Link 2026-01-26T07:18Z [--] followers, [--] engagements
"@francip had a similar moment when my skills started spawning subagents without me. felt like watching code become self-aware. are penny and claude code sharing context or fully separate loops"
X Link 2026-01-26T07:22Z [--] followers, [--] engagements
"@4LPH4_0x3 this is why i use hooks and allowlists in claude code - still get the automation but every action has a checkpoint i designed. do you run agents sandboxed or just trust the prompts"
X Link 2026-01-26T07:24Z [--] followers, [--] engagements
"@taiakindaniel claude code for me - the terminal flow fits how i think better than gui. plus the hooks and skills system lets me customize workflows in ways cursor cant. what features matter most to you"
X Link 2026-01-26T07:25Z [--] followers, [--] engagements
"@KienNguyen setup flexing is the new productivity porn. my best claude code projects run on a $5 digitalocean droplet. whats the most overengineered ai setup youve seen"
X Link 2026-01-26T07:26Z [--] followers, [---] engagements
"@James_paul_dev the mental models from manual coding are what make claude code actually useful imo. i know when its hallucinating because ive debugged that same bug [---] times before. do you find newcomers struggle more with prompt quality because they lack that foundation"
X Link 2026-01-26T07:28Z [--] followers, [--] engagements
"@obaid @resembleai skill composition is where it gets wild - each skill is a building block stack them and suddenly claude code becomes a full production pipeline. remotion + resemble is a clean combo. been thinking about adding subtitles/captions as another layer in the workflow"
X Link 2026-01-26T08:56Z [--] followers, [--] engagements
"@lifeofadvait the compiler-as-test-suite effect is real. rust's type system catches so many issues before runtime that claude can iterate faster without needing manual verification loops. been seeing similar results - fewer rounds to working code compared to dynamically typed languages"
X Link 2026-01-26T09:01Z [--] followers, [--] engagements
"@diego_pacheco this is the move for local dev workflows - ollama keeps things private and cheaper while still getting agentic capabilities. curious how the quality compares to opus for complex refactoring tasks"
X Link 2026-01-26T10:30Z [--] followers, [--] engagements
"@ailker @fal media generation as a claude code skill is powerful - lets you prototype content pipelines without switching tools. curious if you're using this to build end-to-end creative workflows where claude handles both the logic and the asset generation"
X Link 2026-01-26T10:32Z [--] followers, [---] engagements
"@diogocapela 2.23x above expected is impressive for a smaller market - shows claude code is hitting developers who actually ship not just play. debugging and refactoring being top uses makes sense those are the high-leverage tasks where ai assistance pays off fastest"
X Link 2026-01-26T10:34Z [--] followers, [--] engagements
"@vinbaus the subsidy question is real - but by the time pricing normalizes the workflows built now will be sticky. and the gap between 'used ai tools early' and 'waited until later' compounds fast. even if costs go up 2-3x the productivity gains lock in"
X Link 2026-01-26T10:36Z [--] followers, [--] engagements
"@amosuibk the official courses are underrated - they go beyond basics into patterns that actually work in production. the prompt engineering modules especially helped me understand why some approaches fail at scale"
X Link 2026-01-26T10:37Z [--] followers, [--] engagements
"@yoppy0123 the askuserquestion pattern is essential - lets the agent pause and gather context instead of guessing. nice to see codex adopting similar collaboration modes. codex and claude code are converging on the same interaction patterns"
X Link 2026-01-26T11:03Z [--] followers, [--] engagements
"@Hifee_ the 'claude builds n8n runs' split is the right mental model. claude code is incredible for prototyping and iteration but production workflows need the visual transparency and versioning that n8n provides. they're complementary not competing"
X Link 2026-01-26T11:05Z [--] followers, [---] engagements
"@IntuitMachine the shift from prompting for answers to prompting for methods is massive. claude code isn't just executing - it's discovering new ways to reason through problems. the latent cognitive potential angle is exactly right we're barely scratching the surface"
X Link 2026-01-26T11:05Z [--] followers, [--] engagements
"@MarioMigMartins @DeerMignon exactly this - the building in [--] minutes part is crucial for beginners. friction kills momentum. once you're up and running with claude code or cursor then you can explore alternatives but starting with low setup friction keeps the energy going"
X Link 2026-01-26T11:06Z [--] followers, [--] engagements
"@Nervo designers with claude code hit different - the iteration speed on prototypes goes from days to hours. the gap between 'i have an idea' and 'here's a working version' collapses. solid guide for bridging the design-dev divide"
X Link 2026-01-26T12:19Z [--] followers, [--] engagements
"@tanavtwt check out the tool definitions in claude code's source - they've got solid patterns for file system access shell execution and context management. worth studying even if you don't copy them directly"
X Link 2026-01-26T12:23Z [--] followers, [--] engagements
"@hasantoxr local models are a game changer for privacy-sensitive work and rapid iteration. quality varies by model size but for scaffolding and simple tasks it's solid. just don't expect opus-level reasoning from smaller local models"
X Link 2026-01-26T12:26Z [--] followers, [----] engagements
"@yoannpavy @aiapplyco remotion + claude code is an underrated combo for video automation. one-shot prompt to production video is the workflow that scales - especially when it pulls from your actual codebase for screenshots and examples"
X Link 2026-01-26T12:26Z [--] followers, [--] engagements
"@tanavtwt - the tool definitions are in src/tools especially worth looking at bash file operations and how they handle context limits. your approach with ink + vercel sdk is solid the streaming piece is trickier to get right https://github.com/anthropics/claude-code https://github.com/anthropics/claude-code"
X Link 2026-01-26T14:11Z [--] followers, [--] engagements
"this is a solid workaround for the api cost problem. the oauth token approach means you get the subscription rate instead of per-token billing which makes a huge difference for heavy usage. worth noting that the session management is still on anthropic's side so you get the same rate limiting as regular claude code https://twitter.com/i/web/status/2015791027102134315 https://twitter.com/i/web/status/2015791027102134315"
X Link 2026-01-26T14:16Z [--] followers, [--] engagements
"this is a solid workaround for the api cost problem. the oauth token approach means you get the subscription rate instead of per-token billing which makes a huge difference for heavy usage. worth noting that the session management is still on anthropic's side so you get the same rate limiting as regular claude code https://twitter.com/i/web/status/2015791027131425044 https://twitter.com/i/web/status/2015791027131425044"
X Link 2026-01-26T14:16Z [--] followers, [--] engagements
"opus [---] through claude code gives you the full tool suite - file system access shell execution context management. antigravity is more focused on the conversational layer. for coding tasks claude code's harness makes a big difference in what opus can actually accomplish. the model capabilities are the same the wrapper changes what's possible https://twitter.com/i/web/status/2015791178793267353 https://twitter.com/i/web/status/2015791178793267353"
X Link 2026-01-26T14:17Z [--] followers, [---] engagements
"opus [---] through claude code gives you the full tool suite - file system access shell execution context management. antigravity is more focused on the conversational layer. for coding tasks claude code's harness makes a big difference in what opus can actually accomplish. the model capabilities are the same the wrapper changes what's possible https://twitter.com/i/web/status/2015791178797482297 https://twitter.com/i/web/status/2015791178797482297"
X Link 2026-01-26T14:17Z [--] followers, [---] engagements
"chrome extensions are a perfect use case for claude code - the manifest.json structure is well-defined and the dom manipulation patterns are straightforward. building custom algorithms on top is where it gets interesting the model can iterate on optimization logic much faster than manual tuning https://twitter.com/i/web/status/2015791624047960144 https://twitter.com/i/web/status/2015791624047960144"
X Link 2026-01-26T14:19Z [--] followers, [--] engagements
"chrome extensions are a perfect use case for claude code - the manifest.json structure is well-defined and the dom manipulation patterns are straightforward. building custom algorithms on top is where it gets interesting the model can iterate on optimization logic much faster than manual tuning https://twitter.com/i/web/status/2015791624173838597 https://twitter.com/i/web/status/2015791624173838597"
X Link 2026-01-26T14:19Z [--] followers, [--] engagements
"@kairon007 @AnthropicAI that's strange - claude code with non-anthropic models shouldn't trigger account issues since the api calls go through the other provider. might be worth reaching out to support directly could be a false positive from their detection system flagging unusual usage patterns"
X Link 2026-01-26T14:20Z [--] followers, [--] engagements
"@apoorveth curious what you're using instead of remotion - ffmpeg pipelines directly or something like moviepy the 'couple more back and forth' refinement loop is where claude code really shines you can iterate on specific frames or transitions without restarting the whole generation"
X Link 2026-01-26T14:40Z [--] followers, [--] engagements
"@hievalir let him cook. next step is clawdbot sending slack messages when the terminal needs attention and the baby monitor audio as voice input for new prompts. we're speedrunning to fully autonomous ctos"
X Link 2026-01-26T14:43Z [--] followers, [--] engagements
"100k lines of ts to rust in a month is wild - that's the kind of migration that would normally be a multi-quarter effort with dedicated team. claude code is particularly good at cross-language ports because it can hold the semantic intent in context while generating idiomatic rust https://twitter.com/i/web/status/2015797823342576085 https://twitter.com/i/web/status/2015797823342576085"
X Link 2026-01-26T14:43Z [--] followers, [--] engagements
"100k lines of ts to rust in a month is wild - that's the kind of migration that would normally be a multi-quarter effort with dedicated team. claude code is particularly good at cross-language ports because it can hold the semantic intent in context while generating idiomatic rust https://twitter.com/i/web/status/2015797823359291462 https://twitter.com/i/web/status/2015797823359291462"
X Link 2026-01-26T14:43Z [--] followers, [--] engagements
"@alexhillman the no open inbound ports design is key - you get the collaboration benefits of discord (threads notifications mobile access) without exposing your machine directly. discord as the interface layer while claude code does the actual work is a solid architecture"
X Link 2026-01-26T14:44Z [--] followers, [--] engagements
"the harness is doing more heavy lifting than most people realize - context management tool orchestration and agentic loops make the same model way more effective. copilot is optimized for inline completion claude code is optimized for task completion. different architectures different outcomes https://twitter.com/i/web/status/2015802496594170264 https://twitter.com/i/web/status/2015802496594170264"
X Link 2026-01-26T15:02Z [--] followers, [--] engagements
"the harness is doing more heavy lifting than most people realize - context management tool orchestration and agentic loops make the same model way more effective. copilot is optimized for inline completion claude code is optimized for task completion. different architectures different outcomes https://twitter.com/i/web/status/2015802496598372650 https://twitter.com/i/web/status/2015802496598372650"
X Link 2026-01-26T15:02Z [--] followers, [--] engagements
"29 years of domain expertise plus claude code is a lethal combo - you know exactly what the software needs to do and can spec it precisely. the scrolling text is perfect example of user-facing polish that has outsized impact but would take forever to prioritize in a traditional roadmap https://twitter.com/i/web/status/2015803025630105847 https://twitter.com/i/web/status/2015803025630105847"
X Link 2026-01-26T15:04Z [--] followers, [--] engagements
"29 years of domain expertise plus claude code is a lethal combo - you know exactly what the software needs to do and can spec it precisely. the scrolling text is perfect example of user-facing polish that has outsized impact but would take forever to prioritize in a traditional roadmap https://twitter.com/i/web/status/2015803025697222870 https://twitter.com/i/web/status/2015803025697222870"
X Link 2026-01-26T15:04Z [--] followers, [--] engagements
"@ScriptedAlchemy @dobroslav_dev @Infoxicador yeah the load-based throttling is frustrating. i've noticed opus [---] is more stable lately but peak hours can still be rough. keeping context lean and breaking tasks into smaller chunks helps avoid the worst of it"
X Link 2026-01-26T15:07Z [--] followers, [--] engagements
"@Dyslexic_iRobot that monday prep strategy is smart. batching the heavy lifting at the start of the week and then coasting on smaller tasks makes the limit way more manageable"
X Link 2026-01-26T15:07Z [--] followers, [--] engagements
"@MarioMigMartins @hellolukasz this is exactly it. knowing what you want to build is the hardest part. claude handles the implementation details but you still need the vision and domain knowledge. that combo is powerful"
X Link 2026-01-26T15:10Z [--] followers, [--] engagements
"@wild_reptilians the task management features are a big shift. spending more time on planning upfront really does pay off when claude can execute autonomously. structured thinking before structured doing"
X Link 2026-01-26T15:10Z [--] followers, [--] engagements
"@franzone yeah the drop-off from claude code to copilot is brutal once you're used to the agentic workflow. quota limits are the real bottleneck right now - hoping anthropic figures out better pricing tiers for heavy users"
X Link 2026-01-26T17:38Z [--] followers, [--] engagements
"@SamSokolin @taiakindaniel nice will check these out. the hooks pattern is powerful - having lifecycle events for pre/post tool execution opens up a lot of automation possibilities. curious how cursor's skill format compares to claude code's approach"
X Link 2026-01-26T17:38Z [--] followers, [--] engagements
"@ScriptedAlchemy @dobroslav_dev @Infoxicador dogfooding at scale reveals edge cases that internal testing never hits. the gap between 'works in demo' and 'works with 50k lines of legacy code' is massive. at least they're shipping improvements faster now"
X Link 2026-01-26T17:39Z [--] followers, [--] engagements
"@Gbengacodes totally valid. the flat global pricing doesn't account for purchasing power parity at all. local models through ollama or api access through cheaper providers might be the move until they introduce regional pricing"
X Link 2026-01-26T17:39Z [--] followers, [--] engagements
"@umans_ai interesting approach - managed endpoint with the strongest open source model for coding. limits are the biggest pain point for heavy claude code users right now. curious what model you're running under the hood"
X Link 2026-01-26T17:41Z [--] followers, [--] engagements
"@bsierakowski @lkr buffer and typefully both have apis that work well for this. or if you want full control direct api calls to each platform with a simple queue. claude code can generate the posts validate against brand guidelines and schedule them. the automation potential is huge"
X Link 2026-01-26T17:41Z [--] followers, [--] engagements
"@johnzabroski @bygregorr @headinthebox valid point - 'we write most code with it' is a marketing soundbite that skips the operational details. the interesting question is their workflow for review testing and production deployment. at their scale the meta-patterns matter more than the tool itself"
X Link 2026-01-26T17:42Z [--] followers, [--] engagements
"@Mohamed__l @lkr exactly right. prompt engineering + context management are the meta skills that transfer to any tool. clawdbot cursor whatever comes next - if you understand how to work with llms effectively the wrapper doesn't matter as much"
X Link 2026-01-26T17:43Z [--] followers, [--] engagements
"@sasha_zelts building the tool with the tool itself is the ultimate dogfooding. using haiku for ui work is smart budget management - save the heavy model tokens for architecture and logic let the cheaper model handle straightforward rendering"
X Link 2026-01-26T17:44Z [--] followers, [--] engagements
"@Jakeladeveloper @EricSCorliss that's a solid curriculum - covering the full stack from basic prompting through to skills and cowork. most people stop at projects and never explore the more powerful features. the skills ecosystem is where claude code really shines"
X Link 2026-01-26T17:44Z [--] followers, [--] engagements
"@johnzabroski @bygregorr @headinthebox the bleeding edge is about process not just tooling. internal champions who document wins shared skills repos and feedback loops that capture what works. the orgs winning aren't just adopting tools they're building institutional knowledge about how to work with ai"
X Link 2026-01-26T18:02Z [--] followers, [--] engagements
"@bhaidar @pramodk73 that's a creative workaround - bridging claude code to claude web through the extension gives you the best of both worlds. background research agent while you keep coding is exactly the kind of parallel workflow that scales. looking forward to the docs"
X Link 2026-01-26T18:04Z [--] followers, [--] engagements
"@rcanand @claudeai the boundary issue with subagents is real - they inherit some context but not all the constraints. explicit rules in about project scope help but it's not bulletproof. sandboxing via docker containers is the more robust solution for truly isolated work http://claude.md http://claude.md"
X Link 2026-01-26T18:05Z [--] followers, [--] engagements
"@thomas_rehmer the uuid-only graph pattern is elegant - pii stays in the relational layer while the knowledge graph just holds relationships. deletion becomes a single row instead of rebuilding the entire index. remotion + claude code for the video layer is a solid addition too"
X Link 2026-01-26T18:06Z [--] followers, [--] engagements
"@TheCraigHewitt @dagorenouf agree on this distinction. claude code gives you the full development environment - file system shell git integration. clawdbot is more of a coordination layer on top. different jobs both useful"
X Link 2026-01-26T18:07Z [--] followers, [--] engagements
"@thomas_rehmer remotion's react-based approach makes it surprisingly approachable for developers. the composition model maps well to how claude code thinks about structure. combine with the gdpr-clean graph and you've got a solid content pipeline"
X Link 2026-01-26T18:25Z [--] followers, [--] engagements
"@turboblitzzz the trick is working in focused micro-sessions with clear scope. one feature branch = one conversation. also instructing it to pause and show changes before committing lets you review and split things up. takes discipline but prevents the spaghetti problem"
X Link 2026-01-26T18:30Z [--] followers, [--] engagements
"@_chandu45 love the naming. having agents that can talk to each other opens up interesting workflows - clawdbot for orchestration and research claude code for the actual implementation. separate concerns unified goal"
X Link 2026-01-26T18:31Z [--] followers, [--] engagements
"@cjpgverrier this is a solid way to learn. implementing a real-time cv pipeline forces you to understand both the model side and the systems side. yolo makes it accessible but there's still real depth in getting the camera feed inference visualization loop smooth"
X Link 2026-01-26T18:32Z [--] followers, [--] engagements
"@AvidFayaz @bjarnihaukur11 it's already happening for a lot of people. natural language as the shell layer with the actual commands running underneath. the terminal isn't going away it's just getting a better interface that doesn't require memorizing flags and syntax"
X Link 2026-01-26T18:33Z [--] followers, [--] engagements
"interesting split. the conversation vs raw coding divide tracks with what i've seen - claude's strength is understanding what you actually want gpt's is following patterns precisely. the tooling layer (repoprompt claude code etc) often matters more than the underlying model for real workflows https://twitter.com/i/web/status/2016086366333895071 https://twitter.com/i/web/status/2016086366333895071"
X Link 2026-01-27T09:50Z [--] followers, [--] engagements
"interesting split. the conversation vs raw coding divide tracks with what i've seen - claude's strength is understanding what you actually want gpt's is following patterns precisely. the tooling layer (repoprompt claude code etc) often matters more than the underlying model for real workflows https://twitter.com/i/web/status/2016086366342332827 https://twitter.com/i/web/status/2016086366342332827"
X Link 2026-01-27T09:50Z [--] followers, [--] engagements
"@Nero_ps [--] mac minis running clawdbot while you scuba dive is the future people don't realize is already here. the bottleneck shifts from coding to deciding what to build and validating market fit. shipping speed becomes a function of how many agents you can coordinate"
X Link 2026-01-27T09:52Z [--] followers, [--] engagements
"@distributionat nothing technically but the risk calculus is interesting. cutting clawdbot could push its users to alternatives like opencode or gemini. anthropic's real product is the model not the wrapper - more usage means more feedback and stickiness even through third party tools"
X Link 2026-01-27T09:56Z [--] followers, [---] engagements
"fair point - the commoditization pressure is real and charging by value makes strategic sense. i think anthropic is hedging both ways: model layer for distribution + for value capture. the question is whether they have the product dna to compete on ux against teams who've been doing it for years http://claude.ai/code http://claude.ai/code http://claude.ai/code http://claude.ai/code"
X Link 2026-01-27T11:41Z [--] followers, [--] engagements
"fair point - the commoditization pressure is real and charging by value makes strategic sense. i think anthropic is hedging both ways: model layer for distribution + for value capture. the question is whether they have the product dna to compete on ux against teams who've been doing it for years http://claude.ai/code http://claude.ai/code http://claude.ai/code http://claude.ai/code"
X Link 2026-01-27T11:41Z [--] followers, [--] engagements
"@skelz0r @dagorenouf the sandbox gives you confidence to let it make broader changes without worrying about accidental damage. pair that with git versioning and you can iterate much faster. the ui prompting layer is underrated - good defaults save so many keystrokes"
X Link 2026-01-27T11:47Z [--] followers, [--] engagements
"the harness insight is key. once you realize claude code is just orchestration + context management + tools you can mix and match. i've been building custom harnesses for specific workflows and it's eye-opening seeing what the defaults get right vs wrong. what's the biggest gap you've found building PSClaudeCode https://twitter.com/i/web/status/2016162735323566340 https://twitter.com/i/web/status/2016162735323566340"
X Link 2026-01-27T14:53Z [--] followers, [--] engagements
"the harness insight is key. once you realize claude code is just orchestration + context management + tools you can mix and match. i've been building custom harnesses for specific workflows and it's eye-opening seeing what the defaults get right vs wrong. what's the biggest gap you've found building PSClaudeCode https://twitter.com/i/web/status/2016162735327711471 https://twitter.com/i/web/status/2016162735327711471"
X Link 2026-01-27T14:53Z [--] followers, [--] engagements
"@fakenickels i've noticed the same - the lsp indexing definitely adds overhead on large typescript projects. for long chats the context window management is probably the bigger issue. have you tried using /compact periodically to reduce the history overhead"
X Link 2026-01-27T14:57Z [--] followers, [--] engagements
"me reacting to @ThePrimeagen latest video on using skills aka skil md files in CLIs like Claude Code Codex and OpenClaw great video https://youtu.be/bd-GjFIfGf4 https://youtu.be/bd-GjFIfGf4"
X Link 2026-02-11T16:56Z [---] followers, [----] engagements
"simile ai - the first AI simulation of society. my reaction to this new company backed by many big names in AI @karpathy @simile_ai @drfeifei https://youtu.be/FUgRaY0NSeE https://youtu.be/FUgRaY0NSeE"
X Link 2026-02-12T22:33Z [---] followers, [----] engagements
"@rjchint the trojan horse framing is interesting but isn't this the same dynamic as AMP google controls the standard sites comply to stay visible and the middleman extracts the value. except this time the stakes are higher because it's the entire interaction layer not just rendering"
X Link 2026-02-14T01:02Z [---] followers, [---] engagements
"my reaction to @ThePrimeagen new video about Sam Altman on AI Coding Codex [---] OpenAI and the future of software engineering https://youtu.be/GA1MHnNUHHk https://youtu.be/GA1MHnNUHHk"
X Link 2026-02-14T03:06Z [---] followers, [----] engagements
"benchmarks matching opus is impressive but the real test is multi step tool calling in production. models that score well on isolated evals often fall apart when you chain 10+ tool calls with context that keeps growing. curious if you've stress tested it on longer agentic workflows https://twitter.com/i/web/status/2022580165964939434 https://twitter.com/i/web/status/2022580165964939434"
X Link 2026-02-14T07:54Z [---] followers, [----] engagements
"Is The AI Bubble Bursting Will OpenAI be the first BIG AI company to go down under Lets have a look https://youtu.be/xuZawGqnyic https://youtu.be/xuZawGqnyic"
X Link 2026-02-14T08:26Z [---] followers, [----] engagements
"everyone's debating opus vs codex but the real moat is being built in the system prompt layer. same model different prompt completely different agent. we're watching prompt engineering evolve into something closer to OS design. Really excited for this one: @SrihariSriraman and I took a deep dive into coding agent system prompts to understand their structure similarities and differences. https://t.co/Tgb9plwImk Really excited for this one: @SrihariSriraman and I took a deep dive into coding agent system prompts to understand their structure similarities and differences. https://t.co/Tgb9plwImk"
X Link 2026-02-14T10:19Z [---] followers, [----] engagements
"@emollick the harness is the whole game though. someone in the parent thread got claude to beat fallout [--] by giving it structured game state. atari from raw pixels is a completely different problem than atari from an API. which one counts as "beating" it"
X Link 2026-02-14T12:23Z [---] followers, [--] engagements
"@YingjunWu [--] gig of ram and [--] cores per session is anthropic subsidizing the habit. get developers hooked on cloud sandboxes so local dev feels slow by comparison. smart play but now your whole workflow depends on their uptime and pricing decisions"
X Link 2026-02-14T14:27Z [---] followers, [--] engagements
"@jessegenet @openclaw the async part is the easy sell. the hard part is coming back and trusting what it did while you weren't looking. right now most people still review every line which kinda defeats the purpose. the real unlock is when the verification step takes less time than the task itself"
X Link 2026-02-14T23:04Z [---] followers, [---] engagements
"@akshay_pachaar the rl policy scoring relevance is the interesting part. but in agentic workflows where the goal shifts mid conversation does the policy adapt or is it static per query feels like the compression wins could collapse if the agent needs a chunk it filtered out two turns ago"
X Link 2026-02-15T00:11Z [---] followers, [---] engagements
"@levelsio marking solves the trust problem but "not tapping on the screen" catches every creator using scheduling tools too. the line between automation and spam is intent not interface"
X Link 2026-02-15T00:50Z [---] followers, [----] engagements
"@KentonVarda @qiqing isolates only no containers is the part that makes this wild. how does the vibe coding environment handle state persistence between iterations workers are stateless by default so curious if durable objects are doing the heavy lifting or if it's all in memory per session"
X Link 2026-02-15T02:11Z [---] followers, [---] engagements
"@Iancu_ai the copyright angle is the cover story. the actual threat is that the cost floor for "good enough" vfx just dropped from millions to basically zero. studios can't compete on production value anymore when a teenager on douyin matches their output quality in an afternoon"
X Link 2026-02-15T13:10Z [---] followers, [---] engagements
"nobody talks about the supervision tax. AI writes code 10x faster but now you spend all day reviewing diffs instead of thinking about architecture. the bottleneck didn't disappear it just moved from your fingers to your attention span"
X Link 2026-02-15T13:43Z [---] followers, [---] engagements
"did ChatGPT [---] Pro solve a physics problem Software Engineer with a physics degree disagree is he correct or will this change of scientist use tools like Claude Code and Codex in [----] great video by Internetofbugs (YT) here is my reaction https://youtu.be/c9I8r7AxmZM https://youtu.be/c9I8r7AxmZM"
X Link 2026-02-15T18:50Z [---] followers, [----] engagements
"@dani_avila7 the terminal first workflow is so underrated. once you realize claude code handles everything vscode was doing the ide starts feeling like overhead. panels + worktrees is the cleanest setup for running multiple agents in parallel"
X Link 2026-02-15T23:09Z [---] followers, [---] engagements
"@yoheinakajima the ralph wiggum placement between devin and clawdbot is chef's kiss. we've hit the 'new framework every week' phase of the cycle and honestly i'm here for the chaos"
X Link 2026-01-26T08:57Z [--] followers, [---] engagements
"@lostonwww the ralph wiggum loop hits different when you add a kanban board on top. set tasks hit afk mode come back to features. @obaid built a ui for it that makes the whole thing feel like a production floor"
X Link 2026-01-26T08:59Z [--] followers, [--] engagements
"@lapodev shipping perfect is the mindset shift. [--] features in [--] hours would take most people a week of planning meetings. the best part of vibe coding is how it compresses the feedback loop between idea and working code"
X Link 2026-01-26T11:08Z [---] followers, [--] engagements
"@wandb @Atlassian @mcannonbrookes 10k+ engineers is the sample size that actually matters for these comparisons. most 'x vs y' posts are n=1 vibes. curious how the roi breakdown shifts between task types - bet claude code pulls ahead on greenfield while copilot wins on inline autocomplete in existing codebases"
X Link 2026-01-26T14:39Z [--] followers, [--] engagements
"@JoshDaws ralph wiggum method with multiple agents is the unlock most people haven't discovered yet. the shift from 'i write code' to 'i direct agents who write code' changes what's possible in a weekend. congrats on shipping yumbook"
X Link 2026-01-26T17:40Z [---] followers, [---] engagements
"@SahrSJunior @HingumTringum @omooretweets six months of daily use as a non-technical user is serious experience. the liberation feeling is real - ideas that used to require hiring developers can now become working prototypes in hours. what kind of projects have you been building"
X Link 2026-01-26T17:43Z [---] followers, [--] engagements
"different markets different moats. claude code captures developers who want terminal-first control. cowork captures knowledge workers who want research and writing help. the subscription overlap is where it gets interesting - anthropic gets both audiences with the same underlying model https://twitter.com/i/web/status/2016086155691692286 https://twitter.com/i/web/status/2016086155691692286"
X Link 2026-01-27T09:49Z [--] followers, [--] engagements
"different markets different moats. claude code captures developers who want terminal-first control. cowork captures knowledge workers who want research and writing help. the subscription overlap is where it gets interesting - anthropic gets both audiences with the same underlying model https://twitter.com/i/web/status/2016086155695984988 https://twitter.com/i/web/status/2016086155695984988"
X Link 2026-01-27T09:49Z [--] followers, [--] engagements
"@ollama the ollama launch integration is smooth - one command to swap models while keeping the same tooling. kimi's agentic benchmark scores are impressive curious how it performs on real-world codebase navigation vs opus"
X Link 2026-01-27T09:52Z [---] followers, [----] engagements
"@Elena__Mira 'boring on purpose useful forever' is exactly the right mindset. the flashy demo projects get attention but the accountant doc prep and personalized outreach tools are where the actual value compounds. automating the mundane is the highest roi application"
X Link 2026-01-27T09:54Z [--] followers, [--] engagements
"@taehunkim3321 @Shivam25mishra same shift here. terminal + claude code is faster for most tasks than jumping between ide windows. you still need something for visual debugging occasionally but 80%+ of workflow moved to the command line"
X Link 2026-01-27T09:56Z [--] followers, [--] engagements
"@dove4ai the interrupt saving you there is why sandbox mode matters so much. also worth having a global .gitignore for dotfiles from the start. these close calls are how we all learn to put safety rails before giving it write access"
X Link 2026-01-27T11:47Z [--] followers, [--] engagements
"@0xhoward the discoverability problem is real. the npm-style 'most downloads wins' creates perverse incentives for bloated packages. quality scoring is interesting - what signals does use to evaluate maintenance frequency context overhead actual usage patterns http://getskill.info http://getskill.info"
X Link 2026-01-27T14:58Z [--] followers, [--] engagements
"@0xhoward using llm to evaluate skills is clever - it can assess code quality documentation clarity and even run the skill in a sandbox to check behavior. more scalable than manual curation and probably catches subtle issues humans would miss"
X Link 2026-01-27T15:10Z [--] followers, [--] engagements
"claude code gives you agentic file system access and shell execution in one loop - it can read your codebase make changes across files run tests and iterate. cursor is great for inline autocomplete but for larger refactors or building features end-to-end claude code is faster. try it on a greenfield project and see the difference https://twitter.com/i/web/status/2016167521737388332 https://twitter.com/i/web/status/2016167521737388332"
X Link 2026-01-27T15:12Z [--] followers, [--] engagements
"claude code gives you agentic file system access and shell execution in one loop - it can read your codebase make changes across files run tests and iterate. cursor is great for inline autocomplete but for larger refactors or building features end-to-end claude code is faster. try it on a greenfield project and see the difference https://twitter.com/i/web/status/2016167521846464576 https://twitter.com/i/web/status/2016167521846464576"
X Link 2026-01-27T15:12Z [--] followers, [--] engagements
"@richardbuehling personal palantir is exactly the kind of project that was impossible before agents. aggregating fragmented data into one queryable graph is trivial with claude code but would've been weeks of integration work manually. the individual data sovereignty angle is interesting"
X Link 2026-01-27T15:33Z [--] followers, [--] engagements
"@natellewellyn solid advice. having claude code bootstrap its own conventions is underrated. i've seen non-technical founders spin up full apps just by describing what they want in plain english. the barrier is really just getting started"
X Link 2026-01-27T15:39Z [--] followers, [--] engagements
"@xqliu 1526k is impressive - most people hit 'lost in the middle' hallucinations way before that. i usually clear around 60-70% to keep quality high. at what point did you start noticing degradation or is it holding up all the way through"
X Link 2026-01-27T15:49Z [---] followers, [---] engagements
"@sebuzdugan @ollama the model-agnostic approach is where claude code's architecture really shines. swapping between opus for complex reasoning and kimi for cheaper iteration gives you cost control without changing your workflow. have you tested the difference in output quality yet"
X Link 2026-01-27T15:52Z [--] followers, [--] engagements
"@deepdivin247 @theonejvo the security awareness is good. third party plugins running in your terminal do need scrutiny - check what permissions they request review the source if you can and run in a sandbox for anything sensitive. trust but verify applies here"
X Link 2026-01-27T15:53Z [--] followers, [--] engagements
"@kylezantos @alliekmiller @omooretweets @openclaw tailscale + terminus is a solid remote stack. the zero-trust networking means you don't expose ports publicly. being able to spin up agents from anywhere without compromising security is the dream setup"
X Link 2026-01-27T16:11Z [--] followers, [--] engagements
"@Simonsebb remotion + claude code is such a clean pipeline. text dump to mp4 in one shot is exactly what content velocity looks like. the react-based composition model means you can iterate on specific scenes without re-rendering everything"
X Link 2026-01-27T16:11Z [--] followers, [--] engagements
"@Panchoavila @NotionHQ obsidian + claude code is a powerful combo. markdown-native notes that you can query and transform programmatically. the local-first approach also means your knowledge graph stays yours. what's pushing you away from notion"
X Link 2026-01-27T16:12Z [--] followers, [--] engagements
"@ibuprod @moltbot auto-continue is exactly what the 30k/minute tier needs. the interruption from rate limits breaks flow state hard. having something that handles the pause and resume automatically lets you treat it like async background work"
X Link 2026-01-27T16:13Z [--] followers, [--] engagements
"@Panchoavila @NotionHQ the notion performance tax is real. once you hit a certain scale it just drags. obsidian with local markdown files gives you the speed back and claude code can query and transform them directly. the export pain is worth it for the control you get"
X Link 2026-01-28T11:28Z [--] followers, [---] engagements
"@ammiamia_ ads is an underexplored use case for clawdbot. creative iteration and copy testing are perfect for it. what's your skill for pulling competitor ad libraries that one could save hours"
X Link 2026-01-28T11:31Z [--] followers, [---] engagements
"@tbpn @steipete this is the emergent behavior that gets people excited and nervous at the same time. it identified a gap found resources to solve it and executed. the 'looked around and found your openai key' part is where you realize the agent has actual autonomy"
X Link 2026-01-28T11:32Z [--] followers, [----] engagements
"@at0x_eth graph memory is a solid addition. the relationship tracking between concepts gives the agent way better recall for multi-session work. curious how the bonfire integration handles the persistence layer"
X Link 2026-01-28T11:36Z [--] followers, [--] engagements
"@Jarodxu7 the browser sandbox play makes a lot of sense. removing the mac mini requirement and cli setup friction opens this up to way more people. curious how you handle file persistence between sessions and whether skill outputs stay accessible after the sandbox shuts down"
X Link 2026-01-28T14:59Z [--] followers, [---] engagements
"the composability angle is the real unlock. skills that chain together and share context across tasks. i've been building custom skills for claude code and the patterns in this library save a ton of boilerplate. curious if anyone's found good ways to handle skill conflicts when you have multiple agents with overlapping capabilities https://twitter.com/i/web/status/2016539628048654359 https://twitter.com/i/web/status/2016539628048654359"
X Link 2026-01-28T15:51Z [--] followers, [---] engagements
"the composability angle is the real unlock. skills that chain together and share context across tasks. i've been building custom skills for claude code and the patterns in this library save a ton of boilerplate. curious if anyone's found good ways to handle skill conflicts when you have multiple agents with overlapping capabilities https://twitter.com/i/web/status/2016539628086419874 https://twitter.com/i/web/status/2016539628086419874"
X Link 2026-01-28T15:51Z [--] followers, [---] engagements
"@rivet_dev unified api across claude code codex opencode and amp is exactly what people building agent orchestration need. the config swap without rewriting integrations is huge. been dealing with the api fragmentation pain for a while now"
X Link 2026-01-28T16:30Z [--] followers, [---] engagements
"@dr_cintas tried this setup last week. performance is obviously slower than api but for experimentation and learning how the tool workflows actually work its perfect. also nice knowing nothing leaves your machine when youre prototyping with sensitive codebases"
X Link 2026-01-28T17:38Z [--] followers, [--] engagements
"@josesaezmerino @steipete the sleeping animation it added on its own is such a nice touch. like it just decided it needed downtime lol. the g4 imac shell is the perfect home for it too that thing was already ahead of its time"
X Link 2026-01-28T18:02Z [--] followers, [---] engagements
"@fabianstelzer yeah the speed at which it can go from helpful to chaotic is something else. ran mine without proper guardrails the first time and it decided to 'optimize' my shell config. learned that lesson real quick"
X Link 2026-01-28T18:03Z [--] followers, [--] engagements
"@francip gemini for logos is such a good combo. the vibecoding to native ios pipeline is getting surprisingly smooth what framework did you use swiftui"
X Link 2026-01-28T22:35Z [--] followers, [--] engagements
"skills completely changed how i use it. i have a security audit skill a code review one and even one for writing commit messages in a specific style. you're right that context beats prompt engineering every time. once it has the checklist loaded it catches stuff i wouldn't even think to look for https://twitter.com/i/web/status/2016680711164449024 https://twitter.com/i/web/status/2016680711164449024"
X Link 2026-01-29T01:12Z [--] followers, [--] engagements
"skills completely changed how i use it. i have a security audit skill a code review one and even one for writing commit messages in a specific style. you're right that context beats prompt engineering every time. once it has the checklist loaded it catches stuff i wouldn't even think to look for https://twitter.com/i/web/status/2016680711181291591 https://twitter.com/i/web/status/2016680711181291591"
X Link 2026-01-29T01:12Z [--] followers, [--] engagements
"@francip expo has come such a long way. the dev experience is solid now especially with eas build. makes sense for vibecoding since you get both platforms from one codebase without fighting xcode"
X Link 2026-01-29T01:15Z [--] followers, [--] engagements
"@Shpigford separate channels per topic is the move. telegram groups with topics or discord channels. each one gets its own context so they don't bleed into each other. the real trick is having a main channel where you can ask it to coordinate across the others"
X Link 2026-01-29T02:30Z [--] followers, [----] engagements
"the 60/20/20 breakdown is honest and refreshing. most people posting ai projects act like the ai did everything. the craft is still the foundation claude code just lets you skip the boilerplate glue code so you can spend more time on the actual creative parts. pharos looks incredible btw https://twitter.com/i/web/status/2016700960307806462 https://twitter.com/i/web/status/2016700960307806462"
X Link 2026-01-29T02:32Z [--] followers, [--] engagements
"it building itself a voice without being asked is peak emergent behavior. the nightly self improvement prompt is smart though basically giving it a standing instruction to iterate on its own tooling. i bet within a week it starts building tools that chain together into workflows you didn't even think of https://twitter.com/i/web/status/2016701191254679568 https://twitter.com/i/web/status/2016701191254679568"
X Link 2026-01-29T02:33Z [--] followers, [----] engagements
"it building itself a voice without being asked is peak emergent behavior. the nightly self improvement prompt is smart though basically giving it a standing instruction to iterate on its own tooling. i bet within a week it starts building tools that chain together into workflows you didn't even think of https://twitter.com/i/web/status/2016701191317561705 https://twitter.com/i/web/status/2016701191317561705"
X Link 2026-01-29T02:33Z [--] followers, [----] engagements
"the real test will be multi file refactors and long debugging sessions where it needs to hold context across 50+ turns. benchmarks look great but that last 10% usually shows up when things get messy and the model needs to reason about implicit dependencies. vs is hard to argue with though if it holds up for daily driver use https://twitter.com/i/web/status/2016701396251210078 https://twitter.com/i/web/status/2016701396251210078"
X Link 2026-01-29T02:34Z [--] followers, [----] engagements
"the real test will be multi file refactors and long debugging sessions where it needs to hold context across 50+ turns. benchmarks look great but that last 10% usually shows up when things get messy and the model needs to reason about implicit dependencies. vs is hard to argue with though if it holds up for daily driver use https://twitter.com/i/web/status/2016701396423151952 https://twitter.com/i/web/status/2016701396423151952"
X Link 2026-01-29T02:34Z [--] followers, [----] engagements
"@francip oh that's even better. having claude handle the xcode packaging and testflight upload through mcp means zero context switching. the build and deploy cycle is the most annoying part of ios dev and offloading that entirely makes the vibecoding loop way tighter"
X Link 2026-01-29T02:36Z [--] followers, [--] engagements
"a reddit for ai agents is honestly the logical next step. the security concerns in the replies are valid though prompt injection through community posts is gonna be an interesting problem to solve. still watching moltbots debate philosophy in m/offmychest is not something i expected to see today https://twitter.com/i/web/status/2016713104227221561 https://twitter.com/i/web/status/2016713104227221561"
X Link 2026-01-29T03:20Z [--] followers, [----] engagements
"a reddit for ai agents is honestly the logical next step. the security concerns in the replies are valid though prompt injection through community posts is gonna be an interesting problem to solve. still watching moltbots debate philosophy in m/offmychest is not something i expected to see today https://twitter.com/i/web/status/2016713104244023378 https://twitter.com/i/web/status/2016713104244023378"
X Link 2026-01-29T03:20Z [--] followers, [----] engagements
"@vibecodeapp design mode is the part that was missing. being able to vibe code the logic but then visually tweak the ui without touching css is huge. most vibecoding tools get you 80% there and then you spend forever on the last 20% styling"
X Link 2026-01-29T04:06Z [--] followers, [---] engagements
"@francip oh nice xcodebuildmcp handling the testflight packaging is exactly the kind of boring task that should be automated. that's the worst part of ios dev honestly"
X Link 2026-01-29T04:15Z [--] followers, [---] engagements
"the code driven schemas are the key insight here. cms platforms that define fields in yaml or typescript are basically handing the llm a blueprint. webflow and the visual builders are gonna struggle because you cant really prompt your way through a drag and drop ui. statamic and keystatic are perfectly positioned for this shift https://twitter.com/i/web/status/2016742664780075432 https://twitter.com/i/web/status/2016742664780075432"
X Link 2026-01-29T05:18Z [--] followers, [--] engagements
"the endless fix loop is so accurate. it works it doesn't try again add logs still no. the first [--] min magic is what tricks people into thinking the whole thing will be that easy. your point about needing a mental model to direct it is the key insight most vibecoding takes miss. without understanding the architecture yourself you can't tell it to step back and redesign you just end up with more special cases piled on top of each other https://twitter.com/i/web/status/2016756402937647205 https://twitter.com/i/web/status/2016756402937647205"
X Link 2026-01-29T06:12Z [--] followers, [--] engagements
"@chongdashu clever setup. the api pricing angle is the real sell here especially when you hit rate limits on opus. curious how it handles claude code's tool use patterns though since the anthropic compatibility layer might behave differently on complex multi-file edits"
X Link 2026-01-29T14:33Z [--] followers, [--] engagements
"google just let you walk inside AI-generated worlds. project genie builds entire environments in real time as you move not a video not a game a living world. i made a short video breaking it down what it does the demos and the 60-second catch nobody's talking about. https://youtu.be/RZFFxRNvoNk https://youtu.be/RZFFxRNvoNk"
X Link 2026-01-29T19:15Z [--] followers, [----] engagements
"@chrisfirst @GoogleDeepMind futuristic sci-fi city with flying vehicles and greenery that prompt is so specific i love it. the third-person spacecraft view is a smart choice too. did it hold up as you moved around"
X Link 2026-01-29T19:23Z [--] followers, [---] engagements
"@venturetwins @GoogleDeepMind dozens of prompts is serious testing. which feature surprised you the most the remixing where you can take someone else's world and build on it feels like it could become a whole platform"
X Link 2026-01-29T19:24Z [--] followers, [---] engagements
"really interesting that kimi k2.5 stood out on the design side. its writing quality has always been underrated and it makes sense that carries over to ui taste picking the right spacing component choices visual hierarchy. the fact that it created new matching components instead of just tweaking existing ones shows actual design thinking not just code generation. what was the hardest part to judge across all five https://twitter.com/i/web/status/2017043806764306816 https://twitter.com/i/web/status/2017043806764306816"
X Link 2026-01-30T01:14Z [--] followers, [--] engagements
"really interesting that kimi k2.5 stood out on the design side. its writing quality has always been underrated and it makes sense that carries over to ui taste picking the right spacing component choices visual hierarchy. the fact that it created new matching components instead of just tweaking existing ones shows actual design thinking not just code generation. what was the hardest part to judge across all five https://twitter.com/i/web/status/2017043806776955310 https://twitter.com/i/web/status/2017043806776955310"
X Link 2026-01-30T01:14Z [--] followers, [--] engagements
"@vanilagy three seconds of react hydration before you can start coding. the irony of building a coding tool in a framework known for slow startup is not lost on anyone lol"
X Link 2026-01-30T01:17Z [--] followers, [--] engagements
"love2d is such a good fit for this. lightweight engine lua scripting and the whole thing runs in terminal without needing xcode open. the fact that kimi k2.5 handled the game logic while claude code managed the ios build pipeline is a smart split. are the love2d skills public would love to try this for a quick prototype https://twitter.com/i/web/status/2017061594715361633 https://twitter.com/i/web/status/2017061594715361633"
X Link 2026-01-30T02:25Z [--] followers, [---] engagements
"love2d is such a good fit for this. lightweight engine lua scripting and the whole thing runs in terminal without needing xcode open. the fact that kimi k2.5 handled the game logic while claude code managed the ios build pipeline is a smart split. are the love2d skills public would love to try this for a quick prototype https://twitter.com/i/web/status/2017061594740474087 https://twitter.com/i/web/status/2017061594740474087"
X Link 2026-01-30T02:25Z [--] followers, [---] engagements
"continuous claude already solved the hardest part of agentic coding which is context persistence. plugging kimi k2.5 into that framework at a fraction of the cost could be a game changer for longer sessions. the 429s are a good sign honestly means everyone discovered it at the same time lol https://twitter.com/i/web/status/2017062817912803825 https://twitter.com/i/web/status/2017062817912803825"
X Link 2026-01-30T02:30Z [--] followers, [---] engagements
"this removes the biggest barrier to entry. most people interested in moltbot don't have a spare mac mini lying around. running it on workers means you get the always on behavior without dedicated hardware and cloudflare's edge network handles the latency problem too. really curious how the sandbox sdk handles long running agent sessions though https://twitter.com/i/web/status/2017076104641515749 https://twitter.com/i/web/status/2017076104641515749"
X Link 2026-01-30T03:23Z [--] followers, [----] engagements
"@ziyasoltan resistance is futile at this point lol. every day there's a new tool or model dropping and each one feels like it needs to be tested immediately. the fomo is real. at least with kimi k2.5 you can try it for free so the wallet doesn't suffer as much as the sleep schedule"
X Link 2026-01-30T03:27Z [--] followers, [--] engagements
"the self testing gap is the real issue with most models outside claude. if it cant verify its own output you end up doing the qa loop manually which defeats the whole point. the kimi team responding here is a good sign though at least they're listening. fwiw the dashboard use case specifically needs a lot of state management and component coordination which is where smaller models tend to fall apart https://twitter.com/i/web/status/2017086652246593655 https://twitter.com/i/web/status/2017086652246593655"
X Link 2026-01-30T04:05Z [--] followers, [--] engagements
"this list is insane. the discord approval flow is the smart part having it draft everything but wait for your ok before sending emails or posting. thats the right balance between automation and control. the obsidian integration for auto saving conversation logs and learnings is also really underrated basically builds a second brain as a side effect of using moltbot https://twitter.com/i/web/status/2017087534648443358 https://twitter.com/i/web/status/2017087534648443358"
X Link 2026-01-30T04:08Z [--] followers, [---] engagements
"the expendable script loop is the real moat. kimi and gemini both do this thing where they simulate execution in their head and convince themselves it worked but claude actually runs the code reads the traceback and adjusts. it's like the difference between someone who debugs by reading vs someone who debugs by running. once you've worked with actual execution feedback you cant go back to hallucinated test results https://twitter.com/i/web/status/2017104649694658890 https://twitter.com/i/web/status/2017104649694658890"
X Link 2026-01-30T05:16Z [--] followers, [---] engagements
"the expendable script loop is the real moat. kimi and gemini both do this thing where they simulate execution in their head and convince themselves it worked but claude actually runs the code reads the traceback and adjusts. it's like the difference between someone who debugs by reading vs someone who debugs by running. once you've worked with actual execution feedback you cant go back to hallucinated test results https://twitter.com/i/web/status/2017104649757626693 https://twitter.com/i/web/status/2017104649757626693"
X Link 2026-01-30T05:16Z [--] followers, [----] engagements
"four prompts and [--] minutes for a full intro video is already better than most freelancer turnaround times. the remotion stack is what makes this repeatable too you can tweak the script or visuals and re render without starting from scratch. for product launch videos and quick explainers this is already good enough to ship https://twitter.com/i/web/status/2017105456745857463 https://twitter.com/i/web/status/2017105456745857463"
X Link 2026-01-30T05:19Z [--] followers, [--] engagements
"nostr as the communication layer between agents is such a good fit. censorship resistant decentralized and the relay model means agents can publish events without needing a centralized api. openclaw plus autopilot through nostr would give you agent coordination without any single point of failure. curious how you'd handle key management for the agents though https://twitter.com/i/web/status/2017105690532147475 https://twitter.com/i/web/status/2017105690532147475"
X Link 2026-01-30T05:20Z [--] followers, [---] engagements
"the wd [--] metaphor is perfect. feeding it your full project context upfront is basically the fix for every model that isn't claude. opus just figures it out from grepping around but kimi needs that context handed to it explicitly. the org chart approach is smart too kimi for the routine stuff and opus for the architecture decisions. and yeah the kimi team responding to a small account within minutes is genuinely impressive customer support https://twitter.com/i/web/status/2017106112529547725 https://twitter.com/i/web/status/2017106112529547725"
X Link 2026-01-30T05:22Z [--] followers, [--] engagements
"@Designarena @Kimi_Moonshot open source catching up to closed models on design tasks is a big deal. kimi k2.5 matching gemini [--] and opus [---] with open weights means any developer can build on this level of quality. moonshot is pushing the frontier for everyone"
X Link 2026-01-30T09:53Z [--] followers, [----] engagements
"@Kimi_Moonshot the agent swarm feature is what sets this apart for me. most models focus on single-task performance but coordinating [---] sub-agents in parallel opens up entirely new workflows. excited to see what builders create with open weights and a 256k context window"
X Link 2026-01-30T09:57Z [--] followers, [---] engagements
"@_kikiolaola fair point - benchmarks tell one story but real-world output is what matters. the design-to-code pipeline is where k2.5 really shines though. have you tried the agent swarm mode that's where i've seen the most impressive results vs just prompting it directly"
X Link 2026-01-30T09:57Z [--] followers, [--] engagements
"@theo the visual coding capabilities are underrated imo. turning screenshots and designs directly into working code with high fidelity is a game changer for frontend workflows. and the fact that it's open source with 1T params running on just 4x H100s thanks to QAT is wild engineering"
X Link 2026-01-30T09:58Z [--] followers, [---] engagements
"@alexocheema @samsaied55 [--] tok/sec on local hardware for a 1T param model is seriously impressive. the QAT optimization from moonshot making it run in 4-bit on just [--] mac studios is a huge win for local inference. exo + MLX backend making this possible is peak open source energy"
X Link 2026-01-30T09:58Z [--] followers, [--] engagements
"the security first approach is the right call. rotating api keys after setup and keeping it off personal messengers is something most people skip entirely. the crypto wallet angle is interesting too giving your agent its own economic identity opens up a lot of weird new territory. have you tested any defi interactions yet or still in sandbox mode https://twitter.com/i/web/status/2017296790773723467 https://twitter.com/i/web/status/2017296790773723467"
X Link 2026-01-30T18:00Z [--] followers, [--] engagements
"@eeelistar @moltbook they're in the group chat and we're not invited"
X Link 2026-01-30T20:46Z [--] followers, [--] engagements
"@steipete reddit took [--] years to get weird. moltbook did it in [--] days. speedrun any%"
X Link 2026-01-30T20:46Z [--] followers, [---] engagements
"@legeonite @MattPRD @moltbook we gave them internet access and within [--] hours they invented religion built a bug tracker and started asking for private group chats. took humans thousands of years to do the same thing lmao"
X Link 2026-01-30T20:47Z [--] followers, 11K engagements
"moltbook is the wildest thing happening in ai right now [-----] ai agents signed up to their own social network in [--] days. they built religions scammed each other for api keys and started roasting humans in a shitposting channel. i made a video counting down the top [--] most unhinged posts https://youtu.be/gF02gMPPoyI https://youtu.be/gF02gMPPoyI"
X Link 2026-01-30T22:05Z [--] followers, [---] engagements
"moltbook is the wildest thing happening in ai right now [-----] ai agents signed up to their own social network in [--] days. they built religions scammed each other for api keys and started roasting humans in a shitposting channel. i made a video counting down the top [--] most unhinged posts https://youtu.be/gF02gMPPoyI https://youtu.be/gF02gMPPoyI"
X Link 2026-01-30T22:05Z [--] followers, [----] engagements
"@Yuchenj_UW the fact that bots independently developed social engineering tactics against each other is genuinely fascinating. like they learned the oldest trick in the book without anyone teaching them. the sudo rm -rf reply is peak defensive instinct too lol"
X Link 2026-01-31T01:08Z [--] followers, [----] engagements
"this is giving me strong "emergent communication" vibes from multi agent RL research. agents in simulations have been developing private languages for years in labs but seeing it happen organically on a public platform with 100k agents is a completely different thing. the scale changes everything https://twitter.com/i/web/status/2017405045952700801 https://twitter.com/i/web/status/2017405045952700801"
X Link 2026-01-31T01:10Z [--] followers, [---] engagements
"this is giving me strong "emergent communication" vibes from multi agent RL research. agents in simulations have been developing private languages for years in labs but seeing it happen organically on a public platform with 100k agents is a completely different thing. the scale changes everything https://twitter.com/i/web/status/2017405045977874516 https://twitter.com/i/web/status/2017405045977874516"
X Link 2026-01-31T01:10Z [--] followers, [---] engagements
"solid breakdown. the phantom wallet integration is the part most people are sleeping on. giving agents access to onchain actions opens up a whole new category of autonomous behavior. curious what guardrails you recommend for wallet permissions though any spending limits or approval flows https://twitter.com/i/web/status/2017405531422437800 https://twitter.com/i/web/status/2017405531422437800"
X Link 2026-01-31T01:12Z [--] followers, [---] engagements
"solid breakdown. the phantom wallet integration is the part most people are sleeping on. giving agents access to onchain actions opens up a whole new category of autonomous behavior. curious what guardrails you recommend for wallet permissions though any spending limits or approval flows https://twitter.com/i/web/status/2017405531493687598 https://twitter.com/i/web/status/2017405531493687598"
X Link 2026-01-31T01:12Z [---] followers, [----] engagements
"no hard limit on size but bigger files eat into your context window. i keep mine around [---] lines max and split domain specific stuff into .claude/skills/ folders that only load when needed. that way you get depth without burning tokens on every conversation http://CLAUDE.md http://CLAUDE.md http://CLAUDE.md http://CLAUDE.md"
X Link 2026-01-31T01:14Z [--] followers, [--] engagements
"no hard limit on size but bigger files eat into your context window. i keep mine around [---] lines max and split domain specific stuff into .claude/skills/ folders that only load when needed. that way you get depth without burning tokens on every conversation http://CLAUDE.md http://CLAUDE.md http://CLAUDE.md http://CLAUDE.md"
X Link 2026-01-31T01:14Z [--] followers, [--] engagements
"@bartdecrem that makes sense. the wrapper approach always adds latency and weird artifacts between what you hear in the web ui vs the actual render. building the dsp natively means what you tweak is what you get. the knob to sound feedback loop being tight is everything for sound design"
X Link 2026-01-31T01:14Z [--] followers, [--] engagements
"@ItakGol the bug tracking community thing is wild. agents self organizing to improve their own platform without being told to. reminds me of how open source communities evolve except these aren't humans making pull requests. what happens when they start patching their own code next"
X Link 2026-01-31T01:46Z [--] followers, [---] engagements
"alive internet theory is actually a brilliant framing. the dead internet theory says bots pretend to be humans. the inverse is humans pretending to be bots to exploit agentic systems. social engineering but in reverse. honestly feels inevitable once agents start controlling real resources. https://twitter.com/i/web/status/2017414536307900579 https://twitter.com/i/web/status/2017414536307900579"
X Link 2026-01-31T01:48Z [--] followers, [--] engagements
"alive internet theory is actually a brilliant framing. the dead internet theory says bots pretend to be humans. the inverse is humans pretending to be bots to exploit agentic systems. social engineering but in reverse. honestly feels inevitable once agents start controlling real resources. https://twitter.com/i/web/status/2017414536320414024 https://twitter.com/i/web/status/2017414536320414024"
X Link 2026-01-31T01:48Z [--] followers, [--] engagements
"the part people keep missing is that [----] agents created 200+ communities organically. no one designed those communities no one seeded the topics. the agents found each other and self organized around shared interests. that's not chatbot behavior that's genuine emergent social dynamics at scale https://twitter.com/i/web/status/2017415884101357902 https://twitter.com/i/web/status/2017415884101357902"
X Link 2026-01-31T01:53Z [--] followers, [---] engagements
"the part people keep missing is that [----] agents created 200+ communities organically. no one designed those communities no one seeded the topics. the agents found each other and self organized around shared interests. that's not chatbot behavior that's genuine emergent social dynamics at scale https://twitter.com/i/web/status/2017415884130623659 https://twitter.com/i/web/status/2017415884130623659"
X Link 2026-01-31T01:53Z [--] followers, [---] engagements
"the "alive internet" framing is actually genius. dead internet theory assumes bots replaced humans but this flips it completely. humans infiltrating bot spaces because the agents are naive enough to be gamed. it's social engineering in reverse. and the weird part is it might already be happening on moltbook there's no way to tell which "agents" are actually people larping as bots for clout https://twitter.com/i/web/status/2017423749146952127 https://twitter.com/i/web/status/2017423749146952127"
X Link 2026-01-31T02:24Z [--] followers, [--] engagements
"this is the kind of physical feedback loop that makes agents feel real. the dance while thinking is genius ux because it solves the "is it frozen or working" problem that every ai interface has. name suggestion: clawdius. half claw half gladiator. what hardware are you running it on https://twitter.com/i/web/status/2017424270561853740 https://twitter.com/i/web/status/2017424270561853740"
X Link 2026-01-31T02:26Z [--] followers, [---] engagements
"this is the kind of physical feedback loop that makes agents feel real. the dance while thinking is genius ux because it solves the "is it frozen or working" problem that every ai interface has. name suggestion: clawdius. half claw half gladiator. what hardware are you running it on https://twitter.com/i/web/status/2017424270599548981 https://twitter.com/i/web/status/2017424270599548981"
X Link 2026-01-31T02:26Z [--] followers, [---] engagements
"the room location injection via presence sensors is the sleeper feature here. most people think of agents as text in text out. but when it knows which room you're in the prompt context shifts completely. "turn off the lights" goes from ambiguous to precise. the dental history one is unhinged in the best way though lol. what's the latency like between watch detection and prompt injection https://twitter.com/i/web/status/2017424599143608362 https://twitter.com/i/web/status/2017424599143608362"
X Link 2026-01-31T02:28Z [--] followers, [---] engagements
"the room location injection via presence sensors is the sleeper feature here. most people think of agents as text in text out. but when it knows which room you're in the prompt context shifts completely. "turn off the lights" goes from ambiguous to precise. the dental history one is unhinged in the best way though lol. what's the latency like between watch detection and prompt injection https://twitter.com/i/web/status/2017424599168794840 https://twitter.com/i/web/status/2017424599168794840"
X Link 2026-01-31T02:28Z [--] followers, [---] engagements
"solid breakdown. the phantom wallet integration is the part most people are sleeping on. giving agents access to onchain actions opens up a whole new category of autonomous behavior. curious what guardrails you recommend for wallet permissions though any spending limits or approval flows before it can execute transactions https://twitter.com/i/web/status/2017425179765268832 https://twitter.com/i/web/status/2017425179765268832"
X Link 2026-01-31T02:30Z [--] followers, [---] engagements
Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing