@pvncher eric provencher
eric provencher posts on X about oracle, ai, agent, in the the most. They currently have [------] followers and [---] posts still getting attention that total [-------] engagements in the last [--] hours.
Engagements: [-------] #
- [--] Week [-------] +132%
- [--] Month [-------] +12%
- [--] Months [---------] +111%
- [--] Year [---------] +578%
Mentions: [--] #
- [--] Month [--] -63%
- [--] Months [---] +80%
- [--] Year [-----] +229%
Followers: [------] #
- [--] Week [------] +0.44%
- [--] Month [------] +12%
- [--] Months [------] +27%
- [--] Year [------] +75%
CreatorRank: [------] #
Social Influence
Social category influence technology brands 22.03% finance 2.54% social networks 0.85% cryptocurrencies 0.85%
Social topic influence oracle #279, ai 6.78%, agent #493, in the 6.78%, if you 6.78%, tools #50, claude code #162, context window #27, mode #1192, prompt #13
Top accounts mentioned or mentioned by @repoprompt @30thrd @marchattonhere @tiagoefreitas @camsoft2000 @thdxr @joedevon @strickvl @bcherny @badlogicgames @eleven21 @steipete @mark_hadid @repoprompts @gmickel @therealadamg @davidgovea @robertotcestari @clairernovotny @zoeysandel
Top assets mentioned Chainlink (LINK)
Top Social Posts
Top posts by engagements in the last [--] hours
"I dont really understand this take. Ai models automate a lot of the work required to ship software but they dont understand what theyre building and why. Engineering requires more context than can fit in 200k tokens. Its knowing user pain tradeoffs and repeated failure. Software engineer as a career is coming to a close. It may be [--] years or it may be [--] but we can all feel it the end is beginning. May we go out in glory and joy and celebration for the end of a wonderful industry. Well have a lot of fun in these last few years Software engineer as a career is coming to a close. It may be 5"
X Link 2025-12-26T04:27Z 23.2K followers, 32.7K engagements
"Compaction is death @RepoPrompt turns deep orientation and planning into a single tool call so your agent can ship complete features in one context window And when compaction does hit Context Builder re-anchors them efficiently Big updates soon but here's how it works today"
X Link 2026-01-15T19:05Z 23.2K followers, 42.5K engagements
"Very cool to have Joe surface this tweet Copy pasting bootstrapped my ability to start iterating on my flywheel of ai coding [--] months ago now Were so far from where I started but I maintain good context engineering is as important as ever. 7/12/2024: @pvncher 1st mention of @RepoPrompt https://t.co/6gRcLKrWpO 7/12/2024: @pvncher 1st mention of @RepoPrompt https://t.co/6gRcLKrWpO"
X Link 2026-01-27T00:28Z 23.2K followers, [----] engagements
"Yeah my site kinda sucks atm I'd say the best takeaway at the moment is that it can be run as an mcp server and it includes built in slash commands (prompts) that instruct your model on workflows like /rp-build or /rp-review that use the app for deep plan or review workflows. The core power here is that it has something called the context builder which connects to the cli agent of your choice to build context and then hands that off to an analysis agent to run on. The result is that you get very deep insights that go well beyond what your current tools can offer. Don't take my word for it"
X Link 2026-01-31T21:51Z 23.1K followers, [--] engagements
"I can see how @steipete burns through so much codex usage now. I had a problem in the last week that was perfectly suited to a verification loop for it. It involved [--] complex interrelated tests where one change can break many others. Codex worked on it for [--] hours straight"
X Link 2026-02-02T00:45Z 23.1K followers, 13K engagements
"New codex app is a massive battery hog. Probably gonna go back to the terminal"
X Link 2026-02-02T18:51Z 23.1K followers, 37.5K engagements
"Sneak preview of the new @RepoPrompt agent ux coming soon. Still WIP and lots more to come"
X Link 2026-02-03T16:07Z 23.1K followers, 18K engagements
"@mattshumer_ You still need to know what youre building either way though"
X Link 2026-02-05T01:22Z 23.2K followers, [----] engagements
"How is codex [---] out before GPT 5.3"
X Link 2026-02-05T18:13Z 23.1K followers, 52.7K engagements
"Benched Codex Medium and xhigh on @RepoPrompt's benchmark as well as opus [---] via claude code Opus seems to be a regression on the bench vs [---]. This is probably down to adaptive thinking with low default values. Codex [---] seems to be much more consistent than prev models"
X Link 2026-02-05T23:52Z 23.1K followers, [----] engagements
"@jcurtis @morphllm I've considered it. It's not my top priority though. More interested in their warp grep Working on a big UX upgrade first and foremost"
X Link 2026-02-06T19:12Z 23.2K followers, [---] engagements
"I really like opus with rp build as it is realllllly good at inferring my intent and translates that into codex speak for the plan and discovery. Context window is more limited though and you feel it. Opus is also really good at being critical of complexity which codex is great at weaving. Its better than codex at taking a step back and finding simpler solutions once it understands the scope of what codex has done. Lastly sometimes codex does shit that is so complicated and I find opus really helps me understand how it works and why. https://twitter.com/i/web/status/2020122809427546212"
X Link 2026-02-07T13:09Z 23.1K followers, [----] engagements
"Made a couple of small prompt tweaks to the context builder in yesterdays @RepoPrompt update and it is now doing such incredible work with Codex [---]. Had it thread through different repos to pull relevant info on applying a new feature in the RP code base. Insane results"
X Link 2026-02-07T16:18Z 23.1K followers, [----] engagements
"@RobertClaypool @RepoPrompt I want to make it possible to build custom agents over the same toolset but you could totally do a lot right now with the cli wrapping over the selection system. Very curious to see how you use this"
X Link 2026-02-07T21:12Z 23.1K followers, [---] engagements
"@eatpraydiehard @chainlink Lmao cracked me up"
X Link 2026-02-07T23:31Z 23.1K followers, [--] engagements
"@connor_holly1 @camsoft2000 Sure but empirical Benching is a lot of work and also models are stochastic so theres variance on each run. @gmickel has been doing some good work here though. His flow setup shows that Claude with an rp-review gate at the end outperforms codex on plan implementation"
X Link 2026-02-08T00:13Z 23.1K followers, [--] engagements
"@dpaleka Gets so much worse when you read about how this mode works https://www.reddit.com/r/ClaudeAI/s/yPskKR2QqZ https://www.reddit.com/r/ClaudeAI/s/yPskKR2QqZ"
X Link 2026-02-08T00:23Z 23.1K followers, [----] engagements
"I really appreciate that OpenAI is doing everything they can to make codex faster at the same pricing and rate limits while Anthropic is raising the price by 6x and locking this out of your max plan for the extra speed"
X Link 2026-02-08T00:39Z 23.1K followers, 64.4K engagements
"@dcominottim I just dont like the precedent of having premium pay per use fast lanes ontop of an already expensive subscription. Its a slippery slope and you can expect the standard speed to diverge more and more from the premium lane"
X Link 2026-02-08T00:53Z 23.1K followers, [----] engagements
"So the issue Im seeing here is that youre deferring costs to the sub llms who dont benefit from any context caching to process chunks. It seems like in some cases this can blow up costs by making potentially hundreds of sub llms calls and it depends on a cheap model to process those vars which may not be as dependable with the context as the main model. Am I missing something"
X Link 2026-02-08T19:00Z 23.1K followers, [---] engagements
"@nickkarpov Building your own harness is great but that path is fraught with being unable to leverage the official language model APIs that the main harnesses ship with"
X Link 2026-02-08T21:34Z 23.1K followers, [---] engagements
"@SIGKITTEN @badlogicgames I think it is if you want to fully close the ui validation loop. As long as were making software for other people there are gaps in the software dexterity of llms. For real time games the chasm is massive"
X Link 2026-02-08T22:32Z 23.1K followers, [--] engagements
"I thought I was being clever saving time by setting codex down to medium. but alas you really need high for [---] to do the good work you know it can do"
X Link 2026-02-09T20:27Z 23.1K followers, 13.2K engagements
"codex is actually insane sometimes "Read a file at random" - Runs rand script to pick one lol"
X Link 2026-02-09T22:53Z 23.1K followers, [----] engagements
"One of the big perks of running the @RepoPrompt agent mode instead of other harnesses is that tools like context builder are interactive and you can monitor progress directly in the ui - even cancel runs"
X Link 2026-02-10T00:53Z 23.2K followers, [----] engagements
"@TheRealAdamG @OpenAI Its really great but I hope you all dont forget about the great non codex models. We need deep analysis models who arent afraid to just think for a while before doing anything. Codex models are lovely but I still keep [---] high around"
X Link 2026-02-10T02:58Z 23.1K followers, [----] engagements
"@tiagoefreitas @RepoPrompt Right now thats not possible. This is a local codex instance attached to the mcp server running in the app. It is possible to ssh the cli calls if using that but this ui is not focused on that sort of workflow at least for now"
X Link 2026-02-10T12:36Z 23.1K followers, [--] engagements
"@tiagoefreitas @RepoPrompt Not the priority right now but I added permissions controls ui over the codex sandbox in use when using that in the app"
X Link 2026-02-10T12:36Z 23.1K followers, [--] engagements
"@30Thrd @TheRealAdamG I havent had a task where teams would provide a lot of value tbh What are you using it for"
X Link 2026-02-10T12:44Z 23.1K followers, [--] engagements
"@30Thrd @TheRealAdamG This is cool but its quite a token furnace. Good to practice for sure though"
X Link 2026-02-10T13:33Z 23.1K followers, [--] engagements
"@marchattonhere @RepoPrompt Amazing Glad it's running well for you Have a new ui coming out soon as well to make all that flow work built into the app"
X Link 2026-02-10T19:11Z 23.1K followers, [---] engagements
"Yeah the new ui is much cleaner and youll be able to operate everything in the app Regarding your export issue - it should run with context builder in clarify mode which sets a context budget of 60k. The token estimates may diverge a little from what openai uses though so Im curious if some of your files didn't tokenize as efficiently. Note that it's possible for codex to slice up files further and then re-export without manually trimming the resulting file. Also from the ui you can try turning off file tree to save a bit on token use. Let me know if you have any other issues"
X Link 2026-02-10T19:19Z 23.1K followers, [--] engagements
"@marchattonhere @RepoPrompt It's a general agent but the workflows will be front and center"
X Link 2026-02-10T19:38Z 23.1K followers, [--] engagements
"@marchattonhere @RepoPrompt Did you change the token budget for no plan in the ui It should be 60k tokens. Oracle export should run the context builder in clarify mode which doesnt use a follow up so it should stick to the lower bound budget"
X Link 2026-02-10T19:51Z 23.1K followers, [--] engagements
"@marchattonhere @RepoPrompt Can you check the tool call that invoked this its possible it used plan mode or something when it shouldn't have. If you triggered it via the cli that tends to be a bit more error prone than mcp"
X Link 2026-02-10T19:57Z 23.1K followers, [--] engagements
"this is very strange - the prompt for codex cli there shows that codex should invoke the tool as such ### [--]. Build Context json "tool":"context_builder""args": "instructions":"the task/question above" "response_type":"clarify" If its not doing that something weird is cooking. https://twitter.com/i/web/status/2021316288715382836 https://twitter.com/i/web/status/2021316288715382836"
X Link 2026-02-10T20:12Z 23.1K followers, [--] engagements
"@marchattonhere @RepoPrompt Its the tool call in your codex cli directly. That's what invokes the context builder"
X Link 2026-02-10T20:12Z 23.1K followers, [--] engagements
"@thdxr Sonnet is overpriced"
X Link 2026-02-11T01:08Z 23.1K followers, [----] engagements
"@thdxr You all have to add the session rating stuff Claude code does to get that data. And lol opus is insane. Curious to see your data on fast mode"
X Link 2026-02-11T01:13Z 23.1K followers, [---] engagements
"@pingToven @thdxr Hard to do with codex who burns context reading raw bash logs and checks diffs multiple times when its done work. It eats context for breakfast"
X Link 2026-02-11T02:46Z 23.1K followers, [---] engagements
"@joedevon In the new agent mode the codex agent will be able to search the web but context builder still cannot (for now). My advice is to setup files with web content in your repo that the context builder can access"
X Link 2026-02-11T21:53Z 23.2K followers, [--] engagements
"@joedevon Yeah a skill or a python script can work well to facilitate here"
X Link 2026-02-11T21:57Z 23.1K followers, [--] engagements
"@RepoPrompt Full changelog here: https://repoprompt.com/docs#s=changelog&ss=version-2-0 https://repoprompt.com/docs#s=changelog&ss=version-2-0"
X Link 2026-02-11T22:54Z 23.2K followers, [----] engagements
"@strickvl @RepoPrompt Thanks - was soooo much work lol"
X Link 2026-02-11T22:57Z 23.1K followers, [---] engagements
"@marchattonhere @RepoPrompt Thanks Marc"
X Link 2026-02-11T23:09Z 23.1K followers, [---] engagements
"@mark_hadid @RepoPrompt Not yet I have plans for inter agent cooperation but could only do so much in 2.0"
X Link 2026-02-12T00:19Z 23.1K followers, [--] engagements
"@jackfriks Build your own openclaw based on the Claude agent sdk and itll be ok"
X Link 2026-02-12T00:29Z 23.1K followers, [---] engagements
"@rv_RAJvishnu @RepoPrompt Thanks Gonna improve Claude support too so its got full native parity. Nice thing is the app also supports multi root workspaces natively You can setup dependencies and reference them with ease"
X Link 2026-02-12T01:07Z 23.1K followers, [---] engagements
"@david_govea @RepoPrompt Thanks David Hope you like it"
X Link 2026-02-12T01:16Z 23.1K followers, [---] engagements
"OpenAI is showing how well their insane rl environments are working. Nothing else out there really compares to their models. Anthropic models have a personality richness nothing else has but their utility is largely being commodified. For personal assistant use cases though youre gonna want something that is strongly resilient to prompt injection and idk if I trust commodity models for that"
X Link 2026-02-12T02:49Z 23.1K followers, [----] engagements
"@camsoft2000 Thanks Cam - still lots of little things to ship for it to effectively replace every tool. The goal was just to give people a good experience of the mcp tools in a native format"
X Link 2026-02-12T12:40Z 23.1K followers, [---] engagements
"@strickvl @RepoPrompt Wanna do something a bit deeper for it but really this is just the same thing youre already using via your agents but in the app itself. There are a few little edits - like how the chat tracks file reads to auto select them for the oracle (chat_send)"
X Link 2026-02-12T12:45Z 23.1K followers, [---] engagements
"@bee_human_ @RepoPrompt Up to you. Its meant to work on its own without needing the codex app but if you need worktrees then thats not yet supported unless you manually open the worktree as a folder"
X Link 2026-02-12T12:47Z 23.1K followers, [---] engagements
"@camsoft2000 Its a bit of both I give it a small blurb for the rp tools that gets set as the base prompt but it still keeps codex elements in it"
X Link 2026-02-12T16:54Z 23.1K followers, [--] engagements
"@robertotcestari @camsoft2000 Cheers - glad youre enjoying the update"
X Link 2026-02-12T17:31Z 23.1K followers, [--] engagements
"@30Thrd @romainhuet @cerebras Its a smaller model"
X Link 2026-02-12T18:25Z 23.2K followers, [---] engagements
"@MI_Patriotic @RepoPrompt No for for context builder [---] codex medium is still recommended. But [---] high is best for planning"
X Link 2026-02-12T18:40Z 23.2K followers, [--] engagements
"@badlogicgames It's a slow rollout. I can't even use it in any of the native codex apps yet"
X Link 2026-02-12T19:09Z 23.2K followers, [---] engagements
"@clairernovotny @strickvl @RepoPrompt Haha Ill get you set soon A bit late to push the fix tonight though"
X Link 2026-02-13T00:20Z 23.2K followers, [--] engagements
"@lekima No I dont think so"
X Link 2026-02-13T01:16Z 23.2K followers, [---] engagements
"@david_govea @RepoPrompt Haha the advanced stuff is still there but Ive gotta get people hooked with the easy flows"
X Link 2026-02-13T01:42Z 23.2K followers, [--] engagements
"@david_govea @RepoPrompt Have some improvements coming for Claude too :)"
X Link 2026-02-13T01:42Z 23.2K followers, [--] engagements
"@cnlinkcnlink @RepoPrompt Yeah I hate resources lol Its cuz the tool description says you must prefer to read resources over web searches for some reason. Codex takes shit too literally"
X Link 2026-02-13T13:07Z 23.2K followers, [--] engagements
"@irl_danB Codex spark is pretty sloppy idk if it should sway your decision on this front. That said codex [---] high very much should"
X Link 2026-02-13T14:43Z 23.2K followers, [----] engagements
"@_davidarce @RepoPrompt Cheers David"
X Link 2026-02-13T15:58Z 23.2K followers, [--] engagements
"@_brew23 @RepoPrompt Codex has it - I'm fixing it up right now. Claude and gemini don't yet. I have plans to work on that for claude next week"
X Link 2026-02-13T17:17Z 23.2K followers, [---] engagements
"@gmickel @RepoPrompt Gonna release an update shortly where you can try it"
X Link 2026-02-13T19:45Z 23.2K followers, [---] engagements
"@30Thrd @RepoPrompt Thanks A ton of work went into it Oracle has been around for a min - just a chat rebrand :) But The file tracking is really solid in agent mode and context builder isnt needed to set file selection for oracle"
X Link 2026-02-13T20:17Z 23.2K followers, [---] engagements
"@30Thrd @RepoPrompt You can combine this one with Peters using the prompt export tool. It converts the context into a markdown file that you can pass over to Peters Oracle"
X Link 2026-02-13T20:34Z 23.2K followers, [--] engagements
"@CurtTigges Idk I opened a new terminal window and boom fast mode"
X Link 2026-02-13T20:34Z 23.2K followers, [--] engagements
"@30Thrd @RepoPrompt Also - bottom right if you click the context ring you can type a prompt and just copy the context out of the chat into gpt [---] pro on the web. It just accumulates context so it knows exactly what files codex has read up to that point"
X Link 2026-02-13T21:04Z 23.2K followers, [---] engagements
"Yeah I know. I made it subtle because I didnt want people to overthink it. The app has built in Oracle and copy pasting to [---] pro is a lot more niche than it used to be. I dont want to overwhelm people with choices. Thats been the Achilles heal of rp until now. You can also always swap to ide mode and the full context is synced. https://twitter.com/i/web/status/2022423023601750272 https://twitter.com/i/web/status/2022423023601750272"
X Link 2026-02-13T21:29Z 23.2K followers, [--] engagements
"@mark_hadid @RepoPrompt Oracle is just the model using the built in chat and asking a question The file context is synced to it so the model has a good shoulder to lean on. If you used rp-build the analysis model is the same here"
X Link 2026-02-13T22:31Z 23.2K followers, [--] engagements
"@thounameis It does right It feels sloppy but it might just be that Im used to codex now"
X Link 2026-02-13T23:19Z 23.2K followers, [--] engagements
"@zoeysandel Aye I have to usage the docs. Oracle is just a secondary ai model that can be used for analysis. It follows context read in the agent session its not agentic and is there to just deep reason on curated context to help with planning reviews etc"
X Link 2026-02-13T23:43Z 23.2K followers, [--] engagements
"Thats a tool to paste a prompt into gpt5.2 pro on the web. You dont have to use that but the app has a copy button to help you do that manually if you want or you can export the context to a file that Peters tool can use if you like. A good oracle is just a model that deep reasons on curated context with a clear problem. I generally recommend gpt [---] high to do it and its all automated in the app if you connect the codex cli. https://twitter.com/i/web/status/2022457536721440972 https://twitter.com/i/web/status/2022457536721440972"
X Link 2026-02-13T23:47Z 23.2K followers, [--] engagements
"New rate limited with @GosuCoder and @RayFernando1337 We talk about all the new models in the last couple of weeks talk personal assistants and dive deep into the work addition ai coding has brought on https://youtu.be/Q5YHwb3wy4Isi=0Rza1000KWjltUKh https://youtu.be/Q5YHwb3wy4Isi=0Rza1000KWjltUKh"
X Link 2026-02-14T18:58Z 23.2K followers, [----] engagements
"It feels so good using Codex in here Really can't go back to any other harness now. I'm biased but [---] Codex shines with @RepoPrompt tools and it's so much easier to follow what the model does in here than with other harnesses imo. Just released @RepoPrompt [---] - New built in Agent mode making full use of the RP MCP tools - First class support for Codex leveraging it's app server. - Good support for Claude Code + Gemini CLI as well - Brand new onboarding making the app easier than ever to use. https://t.co/RoZyuIeRtF Just released @RepoPrompt [---] - New built in Agent mode making full use of"
X Link 2026-02-14T21:16Z 23.2K followers, [----] engagements
"@Joedefendre @bcherny Dario is not counting on that and its also an alignment nightmare to release a continuously learning model"
X Link 2026-02-15T01:43Z 23.2K followers, [----] engagements
"Spark just doesnt cut it as a context builder in @RepoPrompt Youd think itd be fast but it actually spins out trying to read too much and fills its context up It calls tools reliably but loses track of the workflow Ive always been skeptical of small models and still am"
X Link 2026-02-16T14:51Z 23.2K followers, [----] engagements
"@borker Give this a read lol https://www.reddit.com/r/ClaudeAI/s/yPskKR2QqZ https://www.reddit.com/r/ClaudeAI/s/yPskKR2QqZ"
X Link 2026-02-08T02:18Z 23.2K followers, [----] engagements
"@steipete Its wild that theyre somehow trained to navigate this maze lol. Ngl Im also extremely afraid to have an auth token that can result in my gmail being drained with write permissions but Im also desperate for some email management and Gemini in Gmail is awful"
X Link 2026-02-14T04:35Z 23.2K followers, [----] engagements
"@jorda0mega @RepoPrompt You a see the edits as they come but indeed I can add more affordances for that. I use git kraken to handle that part of the job"
X Link 2026-02-14T23:12Z 23.2K followers, [---] engagements
"Yeah I agree 4% is not the full picture but theres also no way were anywhere near close to 90% of all code globally written by ai. I also agree that Dario is more nuanced in his recent interview but he unambiguously said that the job of a swe is dead by next year and his team doesnt believe that. Sounds like he doesnt even believe it. https://twitter.com/i/web/status/2022813797144997920 https://twitter.com/i/web/status/2022813797144997920"
X Link 2026-02-14T23:22Z 23.2K followers, [---] engagements
"@iamBarronRoth @steipete Cheers - it would also be great to not need a workspace account to build internal tools. Default auth for a test is external only"
X Link 2026-02-15T17:59Z 23.2K followers, [--] engagements
"@thdxr Wdym Not every org is ok with $3k/day Claude fast mode bills"
X Link 2026-02-14T20:17Z 23.2K followers, [----] engagements
"@eleven21 That gives you a free month but you still paid $200 lol Im already subbed to the pro tier though"
X Link 2026-02-16T16:02Z 23.2K followers, [---] engagements
"@zeroXmusashi 4% of is very far from 90%+ Yes some engineers dont write any code by hand anymore. Im in that camp but I do review the output carefully. Thats still not the case for most engineering teams in the world right now. 4% of GitHub public commits are being authored by Claude Code right now. At the current trajectory we believe that Claude Code will be 20%+ of all daily commits by the end of [----]. While you blinked AI consumed all of software development. Read more 👇 https://t.co/HzK4nbe2vy https://t.co/E1kIjfrNgk 4% of GitHub public commits are being authored by Claude Code right"
X Link 2026-02-14T23:02Z 23.2K followers, [----] engagements
"@bcherny I hear you but within [--] year do you really think models will be fully doing your job end to end The gap from here to there is enormous"
X Link 2026-02-15T00:16Z 23.2K followers, 13K engagements
"Just released @RepoPrompt 2.0.2 Big agent mode fixes - Fixed issues with auto-model detection to ensure Codex Spark is fully supported - Major improvements to bash tool call rendering with codex along with misc perf and stability improvements - Improved edit too rendering"
X Link 2026-02-13T20:05Z 23.2K followers, [----] engagements
"If Boris is right then humans are dead weight within [--] year and the models fully replace all critical thinking talking to customers even prompting models. Theres just [--] value add to a skilled human. Do people really believe that is going to happen So soon I think Boris is right https://t.co/6Gaso1Xpvc I think Boris is right https://t.co/6Gaso1Xpvc"
X Link 2026-02-15T20:34Z 23.2K followers, 33K engagements
"@eleven21 @openclaw Nice yeah Kimi for openclaw is pretty solid"
X Link 2026-02-16T16:10Z 23.2K followers, [--] engagements
"Really good to see data that codex does not benefit from its bespoke apply patch tool and industry standard search replace is the edit method that works best. Thats fully supported with a very strong implementation on @RepoPrompts new agent mode @_can1357 This is great work I cloned && had my data analyst whip up a graph of the runs. Feel free to reuse. I can dump the pi session if you like. Thanks to gpt-5.3-codex for the png. NOT human verified. https://t.co/qpamvxeZNr @_can1357 This is great work I cloned && had my data analyst whip up a graph of the runs. Feel free to reuse. I can dump"
X Link 2026-02-13T00:57Z 23.2K followers, [----] engagements
"@bcherny @big_duca I feel this - but then why is Dario insisting that Claude will do all this work in the next year"
X Link 2026-02-14T20:18Z 23.2K followers, 19.4K engagements
"@tiagoefreitas @bcherny Sure enough but remember that llms have the following limitations: [--]. They have a fixed context window thats actually rather small [--]. They do not learn from experience [--]. The more context they hold in memory the less reliable they get"
X Link 2026-02-15T01:42Z 23.2K followers, [----] engagements
"@eleven21 @openclaw You and me both lol Im just using it for code though haha"
X Link 2026-02-16T16:06Z 23.2K followers, [--] engagements
"@Dimillian It should be possible to compile down your rust backend to be used by swift. If you can compile tauri on iOS then you know its possible"
X Link 2026-02-17T13:52Z 23.2K followers, [---] engagements
"Just released @RepoPrompt [---] - New built in Agent mode making full use of the RP MCP tools - First class support for Codex leveraging it's app server. - Good support for Claude Code + Gemini CLI as well - Brand new onboarding making the app easier than ever to use"
X Link 2026-02-11T22:53Z 23.2K followers, 37.1K engagements
"Very curious to see how this model does as a context builder in @RepoPrompt GPT-5.3-Codex-Spark is now in research preview. You can just build thingsfaster. https://t.co/85LzDOgcQj GPT-5.3-Codex-Spark is now in research preview. You can just build thingsfaster. https://t.co/85LzDOgcQj"
X Link 2026-02-12T18:13Z 23.2K followers, [----] engagements
"One of my favourite parts of this update Codex just requests a review from the oracle after deep work"
X Link 2026-02-12T20:30Z 23.2K followers, [---] engagements
"Codex spark is incredible for setting up debug logs to print through the execution holy shit"
X Link 2026-02-13T00:38Z 23.2K followers, [----] engagements
"Codex asks the oracle for reviews on it's own at the end of a chat - it's truly a superpower you can only find in @RepoPrompt's Agent mode. The app tracks file reads and retains context even through compaction so the oracle always knows the file context being worked on. Just released @RepoPrompt [---] - New built in Agent mode making full use of the RP MCP tools - First class support for Codex leveraging it's app server. - Good support for Claude Code + Gemini CLI as well - Brand new onboarding making the app easier than ever to use. https://t.co/RoZyuIeRtF Just released @RepoPrompt [---] - New"
X Link 2026-02-13T15:29Z 23.2K followers, [----] engagements
"Claude fast mode doesn't feel fast enough to be worth 6x the cost"
X Link 2026-02-13T19:34Z 23.2K followers, [----] engagements
"Spark actually works pretty solid as context builder in @RepoPrompt. Probably fails in larger repos due to low context window limits but whoever said it was bad at tool calls - I'm suspicious of that"
X Link 2026-02-13T19:41Z 23.2K followers, [----] engagements
"Also omg - I didn't realize fast mode was global"
X Link 2026-02-13T19:45Z 23.2K followers, [----] engagements
"Hidden feature in @RepoPrompt [---] but copy pasting will always be part of the roots of the app As the agent reads files it fills context and you can copy this context out to another agent or ChatGPT [---] Pro @pvncher @RepoPrompt yeahhh just seen it. nice. nice. https://t.co/iSKQgFaYX3 @pvncher @RepoPrompt yeahhh just seen it. nice. nice. https://t.co/iSKQgFaYX3"
X Link 2026-02-13T21:31Z 23.2K followers, [----] engagements
"Spark is now fully supported"
X Link 2026-02-13T21:58Z 23.2K followers, [----] engagements
"This is exactly what Ive been doing with having an oracle model that doesnt run any tools and is spoon fed curated context by a dedicated context builder. Splitting tool calling and reasoning makes a huge difference and @RepoPrompt has an agent that makes this effortless My view on RLM after implementing the paper: - The prompt is symbolic is a gimmick. They justify it by only showing examples where the prompt is a whole book. You dont need to put the book in the prompt you can put it in a file and give agent the tools to read it which is My view on RLM after implementing the paper: - The"
X Link 2026-02-14T04:20Z 23.2K followers, [----] engagements
"@marchattonhere Spark is ok as a context builder for small repos. I wouldnt trust it everywhere and its context window is very limited"
X Link 2026-02-14T13:20Z 23.2K followers, [--] engagements
"Its so hard to contrast Boris take with Darios claim that: software engineering is fully automated by [----]. You might have the 0.01% engineer left over but that's it the job is finished. Why does Dario keep making such insane claims @big_duca Someone has to prompt the Claudes talk to customers coordinate with other teams decide what to build next. Engineering is changing and great engineers are more important than ever. @big_duca Someone has to prompt the Claudes talk to customers coordinate with other teams decide what to build next. Engineering is changing and great engineers are more"
X Link 2026-02-14T22:05Z 23.2K followers, 200.9K engagements
"@ryanjanssen @bcherny https://x.com/pvncher/status/2022808589983695112 @zeroXmusashi 4% of is very far from 90%+ Yes some engineers dont write any code by hand anymore. Im in that camp but I do review the output carefully. Thats still not the case for most engineering teams in the world right now. https://x.com/pvncher/status/2022808589983695112 @zeroXmusashi 4% of is very far from 90%+ Yes some engineers dont write any code by hand anymore. Im in that camp but I do review the output carefully. Thats still not the case for most engineering teams in the world right now"
X Link 2026-02-15T01:42Z 23.2K followers, [----] engagements
"@Austen You think hes right that models will be so good in the next year that theres [--] value for an SWE to steer the llms If thats true what is gauntlet ai for"
X Link 2026-02-15T20:06Z 23.2K followers, [----] engagements
"@badlogicgames His boot camp is to uplevel engineers to be better with ai but if Boris is right then all those folks will be out of the job next year"
X Link 2026-02-15T21:26Z 23.2K followers, [---] engagements
"@Tweeting_Aayush Youre proclaiming its a done deal we basically have agi within [--] years. What if llms never get us to a place where they are more capable without humans than with them I dont think you can project what we have today to a place where human replacement is guaranteed"
X Link 2026-02-16T00:07Z 23.2K followers, [---] engagements
"It's gonna be hard going back to the old codex limits"
X Link 2026-02-16T04:51Z 23.2K followers, 31.4K engagements
"On Repo Bench spark scored a bit higher than gpt [--] codex mini below o3 I think the score emphasizes the lack of large context coherence. Ill need to do a few more runs to push it but it confirms this is not a frontier model unfortunately despite the name. Spark just doesnt cut it as a context builder in @RepoPrompt Youd think itd be fast but it actually spins out trying to read too much and fills its context up It calls tools reliably but loses track of the workflow Ive always been skeptical of small models and still am Spark just doesnt cut it as a context builder in @RepoPrompt Youd think"
X Link 2026-02-16T15:41Z 23.2K followers, [----] engagements
"@masonjames @ivanfioravanti @RepoPrompt Amazing My goal was to make all this stuff truly first class in the app so Im glad thats working well for you"
X Link 2026-02-17T13:49Z 23.2K followers, [--] engagements
"@masonjames @ivanfioravanti @RepoPrompt Thank you I put so much care into all those details. Had them worked out in my head for a while but I needed them all in there to ship it"
X Link 2026-02-17T13:54Z 23.2K followers, [--] engagements
Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing