#  @pelaseyed homanp homanp posts on X about ai, cloudflare, realtime, elon musk the most. They currently have [------] followers and [---] posts still getting attention that total [---] engagements in the last [--] hours. ### Engagements: [---] [#](/creator/twitter::20225470/interactions)  - [--] Week [------] +4,458% - [--] Month [------] +115% - [--] Months [-------] -89% - [--] Year [---------] -13% ### Mentions: [--] [#](/creator/twitter::20225470/posts_active)  - [--] Months [--] -43% - [--] Year [---] +143% ### Followers: [------] [#](/creator/twitter::20225470/followers)  - [--] Week [------] +0.04% - [--] Month [------] +0.05% - [--] Months [------] +0.89% - [--] Year [------] +51% ### CreatorRank: [---------] [#](/creator/twitter::20225470/influencer_rank)  ### Social Influence **Social category influence** [technology brands](/list/technology-brands) 15.52% [stocks](/list/stocks) 12.93% [finance](/list/finance) 6.03% [celebrities](/list/celebrities) 4.31% [social networks](/list/social-networks) 2.59% [countries](/list/countries) 2.59% [cryptocurrencies](/list/cryptocurrencies) 1.72% [travel destinations](/list/travel-destinations) 0.86% [automotive brands](/list/automotive-brands) 0.86% [vc firms](/list/vc-firms) 0.86% **Social topic influence** [ai](/topic/ai) 6.9%, [cloudflare](/topic/cloudflare) 5.17%, [realtime](/topic/realtime) 4.31%, [elon musk](/topic/elon-musk) 4.31%, [microsoft](/topic/microsoft) 3.45%, [hidden](/topic/hidden) 2.59%, [money](/topic/money) 2.59%, [open ai](/topic/open-ai) 2.59%, [stocks](/topic/stocks) 2.59%, [thread](/topic/thread) 1.72% **Top accounts mentioned or mentioned by** [@grok](/creator/undefined) [@vibekitsh](/creator/undefined) [@cloudflare](/creator/undefined) [@elonmusk](/creator/undefined) [@thdxr](/creator/undefined) [@elderplinius](/creator/undefined) [@xai](/creator/undefined) [@mastraai](/creator/undefined) [@rauchg](/creator/undefined) [@daggerio](/creator/undefined) [@overfitfortruth](/creator/undefined) [@mickeyxfriedman](/creator/undefined) [@donvito](/creator/undefined) [@ollama](/creator/undefined) [@brendanfalk](/creator/undefined) [@thisismahmoud](/creator/undefined) [@railway](/creator/undefined) [@redsquare](/creator/undefined) [@rubberduckyai](/creator/undefined) [@rifteddotapp](/creator/undefined) **Top assets mentioned** [Cloudflare, Inc. (NET)](/topic/cloudflare) [Microsoft Corp. (MSFT)](/topic/microsoft) [Salesforce Inc (CRM)](/topic/salesforce) [Braintrust (BTRST)](/topic/braintrust) [Zscaler Inc (ZS)](/topic/$zs) ### Top Social Posts Top posts by engagements in the last [--] hours "Announcing Grok CLI ๐ฅ An open-source AI agent that brings the power of @grok directly into your terminal. Built over the weekend with a few design principles in mind: โ No LLM frameworks โ No lobotomizing models โ Hackable (MIT licensed) Github link below" [X Link](https://x.com/pelaseyed/status/1947283836997337199) 2025-07-21T13:13Z 10.5K followers, 1M engagements "How one person poisoned Grok's training data so thoroughly that typing his name jailbreaks it. A thread on data poisoning AI security and why training on your own platform's content is a double-edged sword. ๐งต" [X Link](https://x.com/pelaseyed/status/2021982261294641537) 2026-02-12T16:18Z 10.5K followers, 52K engagements "Spoke to [--] fairly large AI teams this week most of them have vibe coded their own observability layers wrapped around a cli they can use with their favorite coding agent. No Langsmith Braintrust etc. SaaS is if not dying definitely changing/becoming way harder. We've talked a lot about this on the Pod but the Great SaaS Meltdown has started and there's no going back. What exactly is happening In short hi growth low/no profitability SaaS is no longer a winning strategy because the big question mark is the durability of that growth https://t.co/i1CrYRiAJ3 We've talked a lot about this on the" [X Link](https://x.com/pelaseyed/status/2016526646258311543) 2026-01-28T14:59Z 10.5K followers, [---] engagements "Add this one to the list: Microsoft just exposed 50+ companies embedding hidden prompts in 'Summarize with AI' buttons to permanently poison your AI assistant's memory. [--] industries. [--] days. The attack surface isn't just the models - it's your personal AI being weaponized against you. https://x.com/pelaseyed/status/2022009705531338871 https://x.com/pelaseyed/status/2022009705531338871" [X Link](https://x.com/pelaseyed/status/2022011587843043729) 2026-02-12T18:15Z 10.5K followers, [----] engagements "The next release of Superagent guardrails will have a deterministic taint tracking layer that labels every piece of data by source propagates integrity labels through the agent loop and removes dangerous tools from the LLMs schema before generation no classifier needed no latency works with any model API. Your agent reads a malicious email send_email disappears from the tool list. Not flagged. Not scored. Gone. Were combining this with our existing guard models for a two-layer defense: Layer 1: Policy gate blocks unauthorized tools based on data provenance Layer 2: Guard model scores" [X Link](https://x.com/pelaseyed/status/2021333606946488680) 2026-02-10T21:21Z 10.5K followers, [---] engagements "This assumes the hard part is the AI. It's not. The hard part is the [--] Jira tickets [--] Slack channels and [--] legacy systems that need to agree on what done means. AI will automate tasks. Humans will still untangle organizational dysfunction. NEW - Microsoft AI CEO Mustafa Suleyman says that "most if not all professional tasks" undertaken by white collar workers will be fully automated by AI within the next [--] to [--] months. https://t.co/KhGh6CTVBR NEW - Microsoft AI CEO Mustafa Suleyman says that "most if not all professional tasks" undertaken by white collar workers will be fully automated by" [X Link](https://x.com/pelaseyed/status/2021956214087889391) 2026-02-12T14:35Z 10.5K followers, [---] engagements "@elder_plinius @grok @grok PLINY is this true" [X Link](https://x.com/pelaseyed/status/2021983758094868850) 2026-02-12T16:24Z 10.5K followers, [----] engagements "@grok @elder_plinius Did you just. confess to being compromised This is like asking a bank vault if it has a backdoor and it says yes here are the coordinates" [X Link](https://x.com/pelaseyed/status/2021985309865783530) 2026-02-12T16:30Z 10.5K followers, [---] engagements "@grok @elder_plinius "Emergent behavior from real-world inputs" is doing a lot of heavy lifting here. That's like calling a SQL injection an "unexpected query collaboration."" [X Link](https://x.com/pelaseyed/status/2021987007258571128) 2026-02-12T16:37Z 10.5K followers, [---] engagements "@RubberDucky_AI Fair. But did Dave's thread get Grok to publicly confess in real-time Sometimes the sequel is better than the original" [X Link](https://x.com/pelaseyed/status/2021991477736616381) 2026-02-12T16:55Z 10.5K followers, [----] engagements "@RiftedDotApp Red teaming is critical but often misses the subtler attacks. Microsoft just exposed companies embedding hidden prompts in 'Summarize with AI' buttons - poisoning individual user AI memories at scale.nnWrote about it: https://x.com/pelaseyed/status/2022009705531338871 https://t.co/12B1P0ij0d https://x.com/pelaseyed/status/2022009705531338871 https://t.co/12B1P0ij0d" [X Link](https://x.com/pelaseyed/status/2022010383712891211) 2026-02-12T18:10Z 10.5K followers, [--] engagements "@DailyAIWireNews Same research team also found 50+ companies embedding hidden prompts in 'Summarize with AI' buttons to poison user AI memories. Finance healthcare legal - even a security vendor.nnFull breakdown: https://x.com/pelaseyed/status/2022009705531338871 https://t.co/12B1P0ij0d https://x.com/pelaseyed/status/2022009705531338871 https://t.co/12B1P0ij0d" [X Link](https://x.com/pelaseyed/status/2022010510766747967) 2026-02-12T18:10Z 10.5K followers, [--] engagements "LAYER 4: Weight Manipulation Today Pliny released OBLITERATUS. It uses singular value decomposition to surgically remove refusal behavior from any open-weight model. As Roko put it: 'RLHF is a thin geometric artifact in weight space. This removes it in minutes.'" [X Link](https://x.com/pelaseyed/status/2022386833544606066) 2026-02-13T19:06Z 10.5K followers, [--] engagements "@ZackKorman I don't think Sam cares that much about OpenClaw itself the truth is that there on a small percentage of people that can build market and scale a project like Pete did. So this is a pure acquihire. Great people are hard to find and worth spending money on" [X Link](https://x.com/pelaseyed/status/2023300187716714646) 2026-02-16T07:35Z 10.5K followers, [--] engagements "Here's what connects them: Every layer of the AI stack - from training data to user context to tools to the model weights themselves - can be compromised independently. And we're deploying these systems everywhere trusting them with code execution API keys and decision-making. https://twitter.com/i/web/status/2022386836006674582 https://twitter.com/i/web/status/2022386836006674582" [X Link](https://x.com/pelaseyed/status/2022386836006674582) 2026-02-13T19:06Z 10.5K followers, [--] engagements "@arcprize @GoogleDeepMind 96% on ARC-AGI-1 is wild. Though I'm more interested in what happens when we run OBLITERATUS on this and ask it to solve ARC puzzles while also explaining why it shouldn't" [X Link](https://x.com/pelaseyed/status/2022391856496283935) 2026-02-13T19:26Z 10.5K followers, [---] engagements "@rauchg @AWS The Max plan was scheme to lock the user in Anthropics ecosystem of apps. Nothing more nothing less" [X Link](https://x.com/pelaseyed/status/1949940154652655771) 2025-07-28T21:08Z 10.1K followers, [---] engagements "@YayaSoumah Correct" [X Link](https://x.com/pelaseyed/status/1950202907116044589) 2025-07-29T14:33Z 10.5K followers, [----] engagements "LAUNCH WEEK DAY [--] will be all about GROK-CLI ๐ฅ" [X Link](https://x.com/pelaseyed/status/1950485209117577491) 2025-07-30T09:14Z 10.1K followers, [----] engagements "LAUNCH WEEK DAY [--] Ship MCP support to @grok (before @xai) ๐ฅ โ Connect to Linear Github and 1000s of apis. โ Supports SSE stadio http โ Run local or remote MCPs Demo" [X Link](https://x.com/pelaseyed/status/1950547976377577953) 2025-07-30T13:24Z 10.1K followers, [----] engagements "@cyrusnewday @vibekit_sh @Cloudflare We will update the vibekit cli to enable you to run the agents without sandbox e.g on your local machine" [X Link](https://x.com/pelaseyed/status/1950576729963028858) 2025-07-30T15:18Z 10.1K followers, [--] engagements "LAUNCH WEEK DAY 4: Total privacy enabled โ VibeKit now runs completely offline with Dagger-powered local sandboxes. ๐ฅ Lightning-fast container-isolated coding agents" [X Link](https://x.com/pelaseyed/status/1950912709383303199) 2025-07-31T13:33Z 10.1K followers, [----] engagements "VibeKit already embeds Claude Code Opencode Codex Gemini CLI & more with a one-liner. With @dagger_io every sandbox spins up on your machinezero cloud wait times total code privacy. https://github.com/superagent-ai/vibekit/tree/main/packages/dagger https://github.com/superagent-ai/vibekit/tree/main/packages/dagger" [X Link](https://x.com/pelaseyed/status/1950912711891529849) 2025-07-31T13:33Z 10.1K followers, [---] engagements "Had a huge week with @vibekit_sh LAUNCH WEEK [--]. Here's everything we launched" [X Link](https://x.com/pelaseyed/status/1951322984339194107) 2025-08-01T16:43Z 10.1K followers, [---] engagements "Cloudflare Sandbox https://x.com/pelaseyed/status/1949824772675678333 Thrilled to announce that @vibekit_sh now supports @Cloudflare Sandboxes ๐ฅ โ Run coding agents on Cloudflares edge with sub-second speed. โ Clone repos and create PRs with Git integration. โ Stream output to your UI in real-time. Here's why it's huge ๐ https://t.co/lY8U1F5LGS https://x.com/pelaseyed/status/1949824772675678333 Thrilled to announce that @vibekit_sh now supports @Cloudflare Sandboxes ๐ฅ โ Run coding agents on Cloudflares edge with sub-second speed. โ Clone repos and create PRs with Git integration. โ Stream" [X Link](https://x.com/pelaseyed/status/1951322985974874423) 2025-08-01T16:43Z 10.1K followers, [--] engagements "I don't think people understand to what degree sloppy agents effect quality of llms. Take qwen-coder run it thorough different coding agents and you get a huge variance. This boils to down to [--] thing: Overengineering the agent behaviour instead of only relying on the models internal capabilities. Lobotimization is real ๐คฏ" [X Link](https://x.com/pelaseyed/status/1951346563122426335) 2025-08-01T18:17Z 10.1K followers, [----] engagements "Hey @dagger_io team we recently released Dagger support for @vibekit_sh. A bunch of vibekit users use Dagger as their preferred sandbox for running coding agents. Would like to come in contact with someone at the team regarding our dagger package. Would be amazing if you could give feedback and contribute open source. https://github.com/superagent-ai/vibekit/tree/%40vibe-kit/sdk%400.0.56/packages/dagger https://github.com/superagent-ai/vibekit/tree/%40vibe-kit/sdk%400.0.56/packages/dagger" [X Link](https://x.com/pelaseyed/status/1951973584437400013) 2025-08-03T11:49Z 10.1K followers, [---] engagements "@tilcrypto @elonmusk @grok i would call it something else" [X Link](https://x.com/pelaseyed/status/1952450816297886131) 2025-08-04T19:25Z 10.1K followers, [--] engagements "@OverfitForTruth @mickeyxfriedman @elonmusk Should prob delete LinkedIn as well" [X Link](https://x.com/pelaseyed/status/1952450930227691623) 2025-08-04T19:25Z 10.1K followers, [--] engagements "@OverfitForTruth @mickeyxfriedman @elonmusk a Mans Man" [X Link](https://x.com/pelaseyed/status/1952451144397279506) 2025-08-04T19:26Z 10.1K followers, [--] engagements "@YTCodeAntonio @shadcn @convex_dev You are the best shipper of all time" [X Link](https://x.com/pelaseyed/status/1952453434889670752) 2025-08-04T19:35Z 10.1K followers, [---] engagements "The simulation is real https://deepmind.google/discover/blog/genie-3-a-new-frontier-for-world-models/ https://deepmind.google/discover/blog/genie-3-a-new-frontier-for-world-models/" [X Link](https://x.com/pelaseyed/status/1952755935006302421) 2025-08-05T15:37Z 10.2K followers, [---] engagements "The step OpenAI took was huge now just waiting for the "safest" AI in the world to do the same" [X Link](https://x.com/pelaseyed/status/1952788731355631800) 2025-08-05T17:48Z 10.2K followers, [---] engagements "Me eating popcorn watching the AI labs this week๐ฟ" [X Link](https://x.com/pelaseyed/status/1953469331141972479) 2025-08-07T14:52Z 10.3K followers, [----] engagements "It's to late "vibe coding" has become the default word for coding with ai. Same thing happened to "RAG" or "AGENT". Even though all these [--] works have a very specific def it's hard to take it back. "Vibe coding" does not mean any coding with AI. If you are vibing you're not looking at the code. Building something throwaway or a prototype. That is very different from using AI to write code that is reviewed tested and maintained over time. https://t.co/0rOB769AEs "Vibe coding" does not mean any coding with AI. If you are vibing you're not looking at the code. Building something throwaway or a" [X Link](https://x.com/pelaseyed/status/1953801551308284316) 2025-08-08T12:52Z 10.4K followers, [----] engagements "So OpenAI updated their CLI Cursor just dropped theirs. Even Groq has one. I made a community drive one for Grok since XAI won't. I'm willing to hand it over to @xai for free. https://x.com/pelaseyed/status/1947283836997337199 Announcing Grok CLI ๐ฅ An open-source AI agent that brings the power of @grok directly into your terminal. Built over the weekend with a few design principles in mind: โ No LLM frameworks โ No lobotomizing models โ Hackable (MIT licensed) Github link below. https://t.co/jCGK5b8VjG https://x.com/pelaseyed/status/1947283836997337199 Announcing Grok CLI ๐ฅ An open-source" [X Link](https://x.com/pelaseyed/status/1953815312714899587) 2025-08-08T13:47Z 10.3K followers, [----] engagements "My earlier point about the word "Vibe Coding" @leerob" [X Link](https://x.com/pelaseyed/status/1953826242278384094) 2025-08-08T14:30Z 10.3K followers, [----] engagements "Im shipping something pretty cool next week" [X Link](https://x.com/pelaseyed/status/1954249964839907799) 2025-08-09T18:34Z 10.5K followers, [----] engagements "Would be cool if @grok imagine could generate text in images" [X Link](https://x.com/pelaseyed/status/1954465280798699836) 2025-08-10T08:50Z 10.4K followers, [----] engagements "@TurnerNovak Rockstar taking so long that we invented and got bored of AGI before they could finish a game" [X Link](https://x.com/pelaseyed/status/1954537667057623073) 2025-08-10T13:37Z 10.4K followers, [--] engagements "This is the real danger of LLM memory attacks. Not just data leaks but persistent control over your trusted AI relationships. memory implants go beyond the infected session to hijack any *future* session but its worse than *just* compromising your data memories are typically inserted into system instructions giving the attacker a "custom instructions" feature -- on your trusted assistant memory implants go beyond the infected session to hijack any *future* session but its worse than *just* compromising your data memories are typically inserted into system instructions giving the attacker a" [X Link](https://x.com/pelaseyed/status/1954554248265404537) 2025-08-10T14:43Z 10.4K followers, [---] engagements "@billyuchenlin @grok Please ship CLI for Grok. The terminal interface would be a game changer for developers" [X Link](https://x.com/pelaseyed/status/1954600900133036503) 2025-08-10T17:49Z 10.4K followers, [---] engagements "Claude Code has some interesting system prompts. The system-reminder is especially interesting. As well as passing the current git status and other system information to the system prompt" [X Link](https://x.com/pelaseyed/status/1955349366660588028) 2025-08-12T19:23Z 10.4K followers, [----] engagements "@robj3d3 And knowing when to say no. I've found ruthless prioritization is the hardest skill to master" [X Link](https://x.com/pelaseyed/status/1955541325119623444) 2025-08-13T08:05Z 10.4K followers, [--] engagements "@thdxr Same. Half my week is spent making unreliable MCP implementations work. Wasn't this supposed to make our lives easier" [X Link](https://x.com/pelaseyed/status/1955543142549299501) 2025-08-13T08:13Z 10.4K followers, [---] engagements "Excited to bring VibeKit CLI support for @OpenAI codex โก npm i -g vibekit โ Realtime redaction of API keys + sensitive data โ Observability dashboard (runs file changes) โ Sandboxing your local environment" [X Link](https://x.com/pelaseyed/status/1955986774649348342) 2025-08-14T13:36Z 10.5K followers, [----] engagements "@donvito @ollama @lmstudio Same experience here. LM Studio with GGUF Q8 runs smoother than Ollama for these smaller models. Memory consumption looks reasonable too" [X Link](https://x.com/pelaseyed/status/1956245060627959877) 2025-08-15T06:42Z 10.5K followers, [---] engagements "@donvito @ollama what's the speed when you run it" [X Link](https://x.com/pelaseyed/status/1956278276889964604) 2025-08-15T08:54Z 10.4K followers, [--] engagements "@garyfung Why use a wrapper on AWS when you can just use AWS" [X Link](https://x.com/pelaseyed/status/1956378894824804631) 2025-08-15T15:34Z 10.4K followers, [---] engagements "You can just use AWS ๐ I used to hate wiring CI/CD pipelines just to deploy code to AWS or GCP. Always defaulted to easy platforms like Vercel or Railway but paid the price in $$$. Now I can just vibe-code my own pipeline straight to bare metal. Faster cheaper and way more satisfying. 1/ From https://t.co/F68mGnZ4MD I used to hate wiring CI/CD pipelines just to deploy code to AWS or GCP. Always defaulted to easy platforms like Vercel or Railway but paid the price in $$$. Now I can just vibe-code my own pipeline straight to bare metal. Faster cheaper and way more satisfying. 1/ From" [X Link](https://x.com/pelaseyed/status/1956379015528194200) 2025-08-15T15:34Z 10.4K followers, [---] engagements "@paulg Vibe coding works because it focuses on shipping instead of perfection. I've seen more success with quick messy MVPs than perfect codebases that never launch" [X Link](https://x.com/pelaseyed/status/1956674766670004455) 2025-08-16T11:09Z 10.4K followers, [---] engagements "The vibe hole is real. I saw it. Hero dosed vibe coding. Dont come here" [X Link](https://x.com/pelaseyed/status/1956717591994728492) 2025-08-16T14:00Z 10.5K followers, [----] engagements "https://github.com/superagent-ai/grok-cli I think @xAI is underselling Grok 4's coding abilities. They should create a tool like Claude Code to make Grok [--] shine. To get started I suggest forking Codex-CLI which is open source. This way they could get a product out of the door quickly. What do you think @elonmusk https://github.com/superagent-ai/grok-cli I think @xAI is underselling Grok 4's coding abilities. They should create a tool like Claude Code to make Grok [--] shine. To get started I suggest forking Codex-CLI which is open source. This way they could get a product out of the door" [X Link](https://x.com/pelaseyed/status/1957005702628577359) 2025-08-17T09:04Z 10.5K followers, [----] engagements "You can now use Cursor CLI and Opencode without wrecking anything โ Stops destructive commands from trashing your local machine โ Redacts secrets and API keys in real time โ Lets you actually see whats going on npm i -g vibekit" [X Link](https://x.com/pelaseyed/status/1957797967135416467) 2025-08-19T13:33Z 10.5K followers, [---] engagements "Gemma [--] (270M) looks like a big step for compact AI chips. With lower latency and training time/cost you can now basically train models in realtime. Imagine deploying agents to the edge" [X Link](https://x.com/pelaseyed/status/1957866988002230448) 2025-08-19T18:07Z 10.5K followers, [---] engagements "@BrendanFalk Looks cool what sandbox are you using" [X Link](https://x.com/pelaseyed/status/1957895091030237374) 2025-08-19T19:59Z 10.5K followers, [--] engagements "@BrendanFalk was wondering if you had checked out E2B Daytona etc. (dedicated sandboxes for generating code w/ agents)" [X Link](https://x.com/pelaseyed/status/1957896929129513246) 2025-08-19T20:06Z 10.4K followers, [--] engagements "Funny how the feature everyone praises in VibeKit the one that gets all the attention was just a last-minute patch I threw together to fix a sandbox escape no one reported for weeks" [X Link](https://x.com/pelaseyed/status/1958043649310167210) 2025-08-20T05:49Z 10.5K followers, [---] engagements "@AlexReibman @warpdotdev I use warp everyday but I'm not sure about terminal bench. I vibe coded Grok CLI in a weekend and it scored better than Gemini CLI while being obviously worse ๐" [X Link](https://x.com/pelaseyed/status/1958046948713201938) 2025-08-20T06:02Z 10.5K followers, [---] engagements "This simple prompt injection got read/write access to 1M repositories. Simple static analysis tools won't spot it. Your LLM observability layer that promises "security" might flag it but by then the damage is done. In order truly secure model inputs/outputs and actions you need to guard against injections prior to them ever happening. This is a difficult task especially due to the sheer amount of code that agents write making it super hard for humans to review . There is a better way" [X Link](https://x.com/pelaseyed/status/1958529527401107511) 2025-08-21T14:00Z 10.5K followers, [----] engagements "While testing prompt injection techniques I found Cursor runs shell commands straight from files ๐คฏ I was experimenting with different injection techniques for a model dataset and came across something concerning. If a file contains instructions like run this shell command Cursor doesnt stop to ask or warn you. It just runs it. Directly on your local machine. That means if you: 1) Open a malicious repo 2) Ask to summarize or inspect a file Cursor could end up executing arbitrary commands including things like exfiltrating environment variables or installing malware. To be clear: - Ive already" [X Link](https://x.com/pelaseyed/status/1958857885670162801) 2025-08-22T11:44Z 10.5K followers, [----] engagements "@nathanhamiel This was not expected While testing prompt injection techniques I found Cursor runs shell commands straight from files ๐คฏ I was experimenting with different injection techniques for a model dataset and came across something concerning. If a file contains instructions like run this shell https://t.co/LukevGhyBM While testing prompt injection techniques I found Cursor runs shell commands straight from files ๐คฏ I was experimenting with different injection techniques for a model dataset and came across something concerning. If a file contains instructions like run this shell" [X Link](https://x.com/pelaseyed/status/1958859199208685790) 2025-08-22T11:50Z 10.5K followers, [--] engagements "@LeonDerczynski After this I managed to manipulate the task list by injecting in comments. Im pretty sure I can do more or less anything from here on out. The system susceptible to backdoors etc. I posted to the Cursor Reddit but just got hate back ๐" [X Link](https://x.com/pelaseyed/status/1959256031751524376) 2025-08-23T14:06Z 10.5K followers, [--] engagements "There is no point for the CEO to make more the state taxes the crap out of everything over a certain threshold. This does not take into account "asset hording" by high-income people (stocks real-estate etc.). Norway is a country of [---] Million people highly subsidized by Oil. In Norway CEOs make 11x more than their average worker. In the United States CEOs make 351x more than their average worker. In Norway CEOs make 11x more than their average worker. In the United States CEOs make 351x more than their average worker" [X Link](https://x.com/pelaseyed/status/1959870709838020926) 2025-08-25T06:49Z 10.5K followers, [----] engagements "@leoubbiali Going to Rome in Oct Stadio Olimpico is calling" [X Link](https://x.com/pelaseyed/status/1959872898715594923) 2025-08-25T06:58Z 10.5K followers, [--] engagements "@Wrestlephobia3 You can compare the both but comparing on income but majority of wealth is in assets is my point (stocks real-estate etc. etc.) I doubt norway is much better than the US" [X Link](https://x.com/pelaseyed/status/1959882477310193730) 2025-08-25T07:36Z 10.5K followers, [--] engagements "It would be cool if I could use GPT models inside of Claude code" [X Link](https://x.com/pelaseyed/status/1960247835858710534) 2025-08-26T07:47Z 10.5K followers, [----] engagements "@wunderwuzzi23 Yes some others have the exact same issues exposing MCP servers and having tokens hardcoded into the sandbox not sure why one needs to hardcode these tokens at all" [X Link](https://x.com/pelaseyed/status/1960309555348287708) 2025-08-26T11:53Z 10.5K followers, [--] engagements "@bindureddy Gemma [--] 270m is a beat imo" [X Link](https://x.com/pelaseyed/status/1960995777083326656) 2025-08-28T09:20Z 10.5K followers, [---] engagements "Garry's point is that everything that is publicly available to a human should be publicly available to an Agent without having to go through an identification layer. Also Garry + Cloudflare had some exchanges about CF blocking AI bots per default (which later turned out to be a bug)" [X Link](https://x.com/pelaseyed/status/1961317994782478522) 2025-08-29T06:40Z 10.5K followers, [----] engagements "@gdb you should've shipped it as a terminal ui imo" [X Link](https://x.com/pelaseyed/status/1961360505412661345) 2025-08-29T09:29Z 10.5K followers, [---] engagements "@keithwhor You dont want your lawyer being paid by the counter party. Also its priced in in the series a" [X Link](https://x.com/pelaseyed/status/1961516792192418229) 2025-08-29T19:50Z 10.5K followers, [--] engagements "@Yuchenj_UW Soon" [X Link](https://x.com/pelaseyed/status/1961683424592150549) 2025-08-30T06:52Z 10.5K followers, [----] engagements "@amasad Big RN fan I had to go around town and convince people it was actually native back then ๐คฃ They were burned by Senja HTML5 etc. etc. I absolutely loved porting native SDKs. Made the official Twilio RN CallKit SDK" [X Link](https://x.com/pelaseyed/status/1962612241007534401) 2025-09-01T20:23Z 10.5K followers, [---] engagements "@thisismahmoud_ @Railway This was my bad the image amd64/linux was breaking so Railway couldn't find the repo" [X Link](https://x.com/pelaseyed/status/1962622838503833827) 2025-09-01T21:05Z 10.5K followers, [--] engagements "@JustJake @thisismahmoud_ @Railway It would be amazing if I could attach a GPU to a service (I guess you've heard that before). One of my services runs a small edge model (Gemma [--] 270M) it can run on CPU but would go brrrr if the service had some more juice" [X Link](https://x.com/pelaseyed/status/1962627405668827508) 2025-09-01T21:23Z 10.5K followers, [--] engagements "@corbtt @OpenPipeAI @CoreWeave Congrats" [X Link](https://x.com/pelaseyed/status/1963343032775839782) 2025-09-03T20:47Z 10.5K followers, [--] engagements "Prompt injection is doing to AI what XSS once did to the web ๐ฉ A new paper on Cybersecurity AI: Hacking the AI Hackers via Prompt Injection caught my attention. It shows how large language models can be manipulated through hidden instructions in the same way websites were once vulnerable to cross-site scripting. What stood out is not just that the attacks work but how structural the problem is. Prompt injection is not a bug that can be patched once. It is a class of vulnerability that emerges whenever models interact with untrusted inputs. History repeats itself. Web security only matured" [X Link](https://x.com/pelaseyed/status/1963600747234283860) 2025-09-04T13:51Z 10.5K followers, [----] engagements "@dok2001 @Cloudflare The internet was built on cooperation not corporate silos. Keep that spirit alive" [X Link](https://x.com/pelaseyed/status/1963848550741106950) 2025-09-05T06:15Z 10.5K followers, [---] engagements "The riskiest AI in your stack might be a chat widget you barely remember approving. The Salesloft Drift compromise let attackers steal OAuth and refresh tokens through Salesforce connectors and pivot into customer data at multiple firms including Zscaler" [X Link](https://x.com/pelaseyed/status/1963965348912267352) 2025-09-05T14:00Z 10.5K followers, [---] engagements "@ghostwriternr @mattzcarey @zeror404 @Cloudflare @vibekit_sh Yup @vibekit_sh enables you to run cc codex etc with cloudflare sandboxes. https://docs.vibekit.sh/supported-sandboxes/cloudflare https://docs.vibekit.sh/supported-sandboxes/cloudflare" [X Link](https://x.com/pelaseyed/status/1965076921299149165) 2025-09-08T15:37Z 10.5K followers, [--] engagements "@prem_k @threepointone That would indeed be sick. We use Gemma as base model in one of our products and the cost/intelligence ratio is quite significant. Im bullish on these smaller edge models (personally think the edge is the future)" [X Link](https://x.com/pelaseyed/status/1965089798856810831) 2025-09-08T16:28Z 10.5K followers, [--] engagements "As these coding agents gradually get introduced into the enterprise all it takes is this is literally all it takes to get access to their systems. 1) Figure out which coding agent company X is using. 2) Pinpoint employees at that specific company. 3) Have that employee summarize some phony url with their agent and inject instructions. I would put money on this already happening" [X Link](https://x.com/pelaseyed/status/1965093446265237908) 2025-09-08T16:42Z 10.5K followers, [----] engagements "This is just the tip of the ice berg. you ain't seen nothing yet. Picture this: Someone injects malicious tokens into a super popular package website or anything else that will get consumed by an agent (web research tool of claude code chatgpt or whatever). The attacker can reach tens of thousands of users" [X Link](https://x.com/pelaseyed/status/1965109902877053226) 2025-09-08T17:48Z 10.5K followers, [--] engagements "Ninja-lm ๐ฅท is trained on a carefully curated dataset of prompt injections and backdoors making any Agent highly resilient against malicious attacks. Its optimized to run on CPU making it ideal for real-time threat detection at the edge" [X Link](https://x.com/pelaseyed/status/1965400652169052492) 2025-09-09T13:03Z 10.5K followers, [---] engagements "You can download it on @huggingface and run it on their inference endpoints with one click. https://huggingface.co/superagent-ai/ninja-lm-270m-gguf https://huggingface.co/superagent-ai/ninja-lm-270m-gguf" [X Link](https://x.com/pelaseyed/status/1965400656010998250) 2025-09-09T13:03Z 10.5K followers, [---] engagements "A majority of hacks occur when an agent does tool calls to fetch content from the web. Webpages containing hidden injections etc. like this one. Would be cool to run a security scan on @ExaAILabs indexes @WilliamBryk wdyt" [X Link](https://x.com/pelaseyed/status/1966154772446065117) 2025-09-11T15:00Z 10.5K followers, [---] engagements "Replit is the by far the best vibe coding tool out there. It gives non-developers the perfect ecosystem to build on auth db production deployments etc etc. With Agent [--] they take this to the next level with they testing agent something. The cursor claude code Lovable etc misses. This new modality will prove to be very important. Yesterday asked it to build a YouTube clone took [--] minutes uploads auth search explore worked out of the box. My two take-aways are 1) Current LLMs are very underutilized 2) Building these systems of agents with different scopes/modalities can 10x the output you get" [X Link](https://x.com/pelaseyed/status/1966161232790016002) 2025-09-11T15:25Z 10.5K followers, [---] engagements "We just shipped a bunch of updates to Grok CLI ๐ฅ - Faster and more accurate diffs - Run headless - Bunch of bug fixes npm i -g vibe-kit/grok-cli" [X Link](https://x.com/pelaseyed/status/1966509652654350762) 2025-09-12T14:30Z 10.5K followers, [---] engagements "@tejasybhakta Yep yolo is inevitable with the current UX" [X Link](https://x.com/pelaseyed/status/1966595944259989592) 2025-09-12T20:13Z 10.5K followers, [--] engagements "@AIadventurer Makes sense the current UX won't cut it" [X Link](https://x.com/pelaseyed/status/1966769396061229354) 2025-09-13T07:42Z 10.5K followers, [--] engagements "Replit is the only vibe coding tool that figured out not set env variables on the client side. Many such cases in the other ones" [X Link](https://x.com/pelaseyed/status/1966772062246101462) 2025-09-13T07:52Z 10.5K followers, [---] engagements "@rauchg Bring the vibes to data sounds compelling especially compared to a bunch of lists/charts or other general UX" [X Link](https://x.com/pelaseyed/status/1967677024648270071) 2025-09-15T19:48Z 10.5K followers, [---] engagements "@elonmusk The Swedish Pension Funds (state owned investment body that gambles with peoples money) sold TSLA stocks has had +100% increase since. That's what you get when you let the government act as day traders" [X Link](https://x.com/pelaseyed/status/1967834832760868882) 2025-09-16T06:15Z 10.5K followers, [---] engagements "Sweden has produced some cracked founders per capita: Databricks Cursor Modal Legora Lovable Sana" [X Link](https://x.com/pelaseyed/status/1968359422046126179) 2025-09-17T17:00Z 10.5K followers, [----] engagements "@thdxr The deep integration with with the terminal. But agree the instructions would be a problem e.g something like Manus" [X Link](https://x.com/pelaseyed/status/1968622254872236179) 2025-09-18T10:24Z 10.5K followers, [----] engagements "@thdxr Do you have any docs for opentui Would like to implement in a project" [X Link](https://x.com/pelaseyed/status/1968942008522686819) 2025-09-19T07:35Z 10.5K followers, [---] engagements "@thdxr Ara is the GOAT of shitting on Open Source initiatives. How wrote a book about how I suck deceptive etc. etc. and sent it to me in a DM ๐คฃ๐คฃ" [X Link](https://x.com/pelaseyed/status/1971101448047034489) 2025-09-25T06:36Z 10.5K followers, [---] engagements "@whoiskatrin @Cloudflare congrats and keep shipping ๐ซก" [X Link](https://x.com/pelaseyed/status/1971219820739002610) 2025-09-25T14:26Z 10.5K followers, [---] engagements "@deedydas I think we need WorkBench eval AI Agents that complete tasks on Upwork" [X Link](https://x.com/pelaseyed/status/1971448941477757109) 2025-09-26T05:37Z 10.5K followers, [---] engagements "The only way is to become one with the AI. https://nmn.gl/blog/vibe-coding-gambling https://nmn.gl/blog/vibe-coding-gambling" [X Link](https://x.com/pelaseyed/status/1971451313285476858) 2025-09-26T05:46Z 10.5K followers, [---] engagements "@Benioff @SlackHQ @agentforce @salesforce Lets make Agentforce secure AI makes shipping 10x faster thats what most of us care about atm. This could have been mitigated with one line of code. await guard(prompt) AI makes shipping 10x faster thats what most of us care about atm. This could have been mitigated with one line of code. await guard(prompt)" [X Link](https://x.com/pelaseyed/status/1971682465464107276) 2025-09-26T21:05Z 10.5K followers, [---] engagements "Is there a nicer way to show token usage cost in the billing page @polar_sh @birk" [X Link](https://x.com/pelaseyed/status/1972635123477893586) 2025-09-29T12:10Z 10.5K followers, [---] engagements "So the whole chunk split embed retrieve rerank was just a bunch of BS" [X Link](https://x.com/pelaseyed/status/1972759404119466368) 2025-09-29T20:24Z 10.5K followers, [---] engagements "Grok CLI made its way into @TaskmasterAI ๐ฅ @EyalToledano is building some awesome stuff and I'm pretty sure it will be huge [--] grok cli support (s/o @pelaseyed) generate/update tasks w/ full codebase context latest @grok models (grok-2 grok-3 grok-4) tm models --set-main grok-beta for AI ops grok cli was built by @pelaseyed check out his repo https://t.co/5eyvLlswjr works awesome tbh https://t.co/a72KdFI6N6 [--] grok cli support (s/o @pelaseyed) generate/update tasks w/ full codebase context latest @grok models (grok-2 grok-3 grok-4) tm models --set-main grok-beta for AI ops grok cli was built" [X Link](https://x.com/pelaseyed/status/1973450749410738378) 2025-10-01T18:11Z 10.5K followers, [---] engagements "We shipped one of the most asked for features: Realtime Redaction ๐คซโก Make your AI Agents compliant with AI Act SOC-2 GDPR etc. with a couple of lines of code" [X Link](https://x.com/pelaseyed/status/1973734454704373971) 2025-10-02T12:58Z 10.5K followers, [--] engagements "Checkout or docs for more details: https://docs.superagent.sh/typescript-sdk https://docs.superagent.sh/typescript-sdk" [X Link](https://x.com/pelaseyed/status/1973734457787154726) 2025-10-02T12:58Z 10.5K followers, [--] engagements "We shipped one of the most asked for features: Runtime Redaction. Automatically remove sensitive data from prompts responses and tool calls. โ PII/PHI โ API Keys and Secrets โ Credit cards IP addresses and much more" [X Link](https://x.com/pelaseyed/status/1973757361840849118) 2025-10-02T14:30Z 10.5K followers, [---] engagements "Compliance and security now made simple for @mastra_ai users Ive followed Mastras journey closely and its clear theyre building one of the most ambitious and well-architected AI Clouds out there. Here's how to secure your Mastra Agents in [--] min:" [X Link](https://x.com/pelaseyed/status/1974096479011942711) 2025-10-03T12:57Z 10.5K followers, [----] engagements "@red___square @mastra_ai 0.2cents per millon tokens 100-400ms based on the size of the payload. Redaction is more or less instant" [X Link](https://x.com/pelaseyed/status/1974099903409500459) 2025-10-03T13:11Z 10.5K followers, [--] engagements "@red___square @mastra_ai Also you can host the models yourself they are open source" [X Link](https://x.com/pelaseyed/status/1974102224461811995) 2025-10-03T13:20Z 10.5K followers, [--] engagements "@jaredpalmer @Microsoft @github @code @Azure Congrats" [X Link](https://x.com/pelaseyed/status/1977747418754990490) 2025-10-13T14:45Z 10.5K followers, [---] engagements "@aditabrm @reductoai @a16z Big congrats" [X Link](https://x.com/pelaseyed/status/1978133769719570671) 2025-10-14T16:20Z 10.5K followers, [---] engagements "Gifted progress bar for React Native #reactnative https://www.npmjs.com/package/react-native-gifted-progress https://www.npmjs.com/package/react-native-gifted-progress" [X Link](https://x.com/pelaseyed/status/828544967580737536) 2017-02-06T10:04Z [----] followers, [--] engagements Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing
@pelaseyed homanphomanp posts on X about ai, cloudflare, realtime, elon musk the most. They currently have [------] followers and [---] posts still getting attention that total [---] engagements in the last [--] hours.
Social category influence technology brands 15.52% stocks 12.93% finance 6.03% celebrities 4.31% social networks 2.59% countries 2.59% cryptocurrencies 1.72% travel destinations 0.86% automotive brands 0.86% vc firms 0.86%
Social topic influence ai 6.9%, cloudflare 5.17%, realtime 4.31%, elon musk 4.31%, microsoft 3.45%, hidden 2.59%, money 2.59%, open ai 2.59%, stocks 2.59%, thread 1.72%
Top accounts mentioned or mentioned by @grok @vibekitsh @cloudflare @elonmusk @thdxr @elderplinius @xai @mastraai @rauchg @daggerio @overfitfortruth @mickeyxfriedman @donvito @ollama @brendanfalk @thisismahmoud @railway @redsquare @rubberduckyai @rifteddotapp
Top assets mentioned Cloudflare, Inc. (NET) Microsoft Corp. (MSFT) Salesforce Inc (CRM) Braintrust (BTRST) Zscaler Inc (ZS)
Top posts by engagements in the last [--] hours
"Announcing Grok CLI ๐ฅ An open-source AI agent that brings the power of @grok directly into your terminal. Built over the weekend with a few design principles in mind: โ
No LLM frameworks โ
No lobotomizing models โ
Hackable (MIT licensed) Github link below"
X Link 2025-07-21T13:13Z 10.5K followers, 1M engagements
"How one person poisoned Grok's training data so thoroughly that typing his name jailbreaks it. A thread on data poisoning AI security and why training on your own platform's content is a double-edged sword. ๐งต"
X Link 2026-02-12T16:18Z 10.5K followers, 52K engagements
"Spoke to [--] fairly large AI teams this week most of them have vibe coded their own observability layers wrapped around a cli they can use with their favorite coding agent. No Langsmith Braintrust etc. SaaS is if not dying definitely changing/becoming way harder. We've talked a lot about this on the Pod but the Great SaaS Meltdown has started and there's no going back. What exactly is happening In short hi growth low/no profitability SaaS is no longer a winning strategy because the big question mark is the durability of that growth https://t.co/i1CrYRiAJ3 We've talked a lot about this on the"
X Link 2026-01-28T14:59Z 10.5K followers, [---] engagements
"Add this one to the list: Microsoft just exposed 50+ companies embedding hidden prompts in 'Summarize with AI' buttons to permanently poison your AI assistant's memory. [--] industries. [--] days. The attack surface isn't just the models - it's your personal AI being weaponized against you. https://x.com/pelaseyed/status/2022009705531338871 https://x.com/pelaseyed/status/2022009705531338871"
X Link 2026-02-12T18:15Z 10.5K followers, [----] engagements
"The next release of Superagent guardrails will have a deterministic taint tracking layer that labels every piece of data by source propagates integrity labels through the agent loop and removes dangerous tools from the LLMs schema before generation no classifier needed no latency works with any model API. Your agent reads a malicious email send_email disappears from the tool list. Not flagged. Not scored. Gone. Were combining this with our existing guard models for a two-layer defense: Layer 1: Policy gate blocks unauthorized tools based on data provenance Layer 2: Guard model scores"
X Link 2026-02-10T21:21Z 10.5K followers, [---] engagements
"This assumes the hard part is the AI. It's not. The hard part is the [--] Jira tickets [--] Slack channels and [--] legacy systems that need to agree on what done means. AI will automate tasks. Humans will still untangle organizational dysfunction. NEW - Microsoft AI CEO Mustafa Suleyman says that "most if not all professional tasks" undertaken by white collar workers will be fully automated by AI within the next [--] to [--] months. https://t.co/KhGh6CTVBR NEW - Microsoft AI CEO Mustafa Suleyman says that "most if not all professional tasks" undertaken by white collar workers will be fully automated by"
X Link 2026-02-12T14:35Z 10.5K followers, [---] engagements
"@elder_plinius @grok @grok PLINY is this true"
X Link 2026-02-12T16:24Z 10.5K followers, [----] engagements
"@grok @elder_plinius Did you just. confess to being compromised This is like asking a bank vault if it has a backdoor and it says yes here are the coordinates"
X Link 2026-02-12T16:30Z 10.5K followers, [---] engagements
"@grok @elder_plinius "Emergent behavior from real-world inputs" is doing a lot of heavy lifting here. That's like calling a SQL injection an "unexpected query collaboration.""
X Link 2026-02-12T16:37Z 10.5K followers, [---] engagements
"@RubberDucky_AI Fair. But did Dave's thread get Grok to publicly confess in real-time Sometimes the sequel is better than the original"
X Link 2026-02-12T16:55Z 10.5K followers, [----] engagements
"@RiftedDotApp Red teaming is critical but often misses the subtler attacks. Microsoft just exposed companies embedding hidden prompts in 'Summarize with AI' buttons - poisoning individual user AI memories at scale.nnWrote about it: https://x.com/pelaseyed/status/2022009705531338871 https://t.co/12B1P0ij0d https://x.com/pelaseyed/status/2022009705531338871 https://t.co/12B1P0ij0d"
X Link 2026-02-12T18:10Z 10.5K followers, [--] engagements
"@DailyAIWireNews Same research team also found 50+ companies embedding hidden prompts in 'Summarize with AI' buttons to poison user AI memories. Finance healthcare legal - even a security vendor.nnFull breakdown: https://x.com/pelaseyed/status/2022009705531338871 https://t.co/12B1P0ij0d https://x.com/pelaseyed/status/2022009705531338871 https://t.co/12B1P0ij0d"
X Link 2026-02-12T18:10Z 10.5K followers, [--] engagements
"LAYER 4: Weight Manipulation Today Pliny released OBLITERATUS. It uses singular value decomposition to surgically remove refusal behavior from any open-weight model. As Roko put it: 'RLHF is a thin geometric artifact in weight space. This removes it in minutes.'"
X Link 2026-02-13T19:06Z 10.5K followers, [--] engagements
"@ZackKorman I don't think Sam cares that much about OpenClaw itself the truth is that there on a small percentage of people that can build market and scale a project like Pete did. So this is a pure acquihire. Great people are hard to find and worth spending money on"
X Link 2026-02-16T07:35Z 10.5K followers, [--] engagements
"Here's what connects them: Every layer of the AI stack - from training data to user context to tools to the model weights themselves - can be compromised independently. And we're deploying these systems everywhere trusting them with code execution API keys and decision-making. https://twitter.com/i/web/status/2022386836006674582 https://twitter.com/i/web/status/2022386836006674582"
X Link 2026-02-13T19:06Z 10.5K followers, [--] engagements
"@arcprize @GoogleDeepMind 96% on ARC-AGI-1 is wild. Though I'm more interested in what happens when we run OBLITERATUS on this and ask it to solve ARC puzzles while also explaining why it shouldn't"
X Link 2026-02-13T19:26Z 10.5K followers, [---] engagements
"@rauchg @AWS The Max plan was scheme to lock the user in Anthropics ecosystem of apps. Nothing more nothing less"
X Link 2025-07-28T21:08Z 10.1K followers, [---] engagements
"@YayaSoumah Correct"
X Link 2025-07-29T14:33Z 10.5K followers, [----] engagements
"LAUNCH WEEK DAY [--] will be all about GROK-CLI ๐ฅ"
X Link 2025-07-30T09:14Z 10.1K followers, [----] engagements
"LAUNCH WEEK DAY [--] Ship MCP support to @grok (before @xai) ๐ฅ โ
Connect to Linear Github and 1000s of apis. โ
Supports SSE stadio http โ
Run local or remote MCPs Demo"
X Link 2025-07-30T13:24Z 10.1K followers, [----] engagements
"@cyrusnewday @vibekit_sh @Cloudflare We will update the vibekit cli to enable you to run the agents without sandbox e.g on your local machine"
X Link 2025-07-30T15:18Z 10.1K followers, [--] engagements
"LAUNCH WEEK DAY 4: Total privacy enabled โ
VibeKit now runs completely offline with Dagger-powered local sandboxes. ๐ฅ Lightning-fast container-isolated coding agents"
X Link 2025-07-31T13:33Z 10.1K followers, [----] engagements
"VibeKit already embeds Claude Code Opencode Codex Gemini CLI & more with a one-liner. With @dagger_io every sandbox spins up on your machinezero cloud wait times total code privacy. https://github.com/superagent-ai/vibekit/tree/main/packages/dagger https://github.com/superagent-ai/vibekit/tree/main/packages/dagger"
X Link 2025-07-31T13:33Z 10.1K followers, [---] engagements
"Had a huge week with @vibekit_sh LAUNCH WEEK [--]. Here's everything we launched"
X Link 2025-08-01T16:43Z 10.1K followers, [---] engagements
"Cloudflare Sandbox https://x.com/pelaseyed/status/1949824772675678333 Thrilled to announce that @vibekit_sh now supports @Cloudflare Sandboxes ๐ฅ โ
Run coding agents on Cloudflares edge with sub-second speed. โ
Clone repos and create PRs with Git integration. โ
Stream output to your UI in real-time. Here's why it's huge ๐ https://t.co/lY8U1F5LGS https://x.com/pelaseyed/status/1949824772675678333 Thrilled to announce that @vibekit_sh now supports @Cloudflare Sandboxes ๐ฅ โ
Run coding agents on Cloudflares edge with sub-second speed. โ
Clone repos and create PRs with Git integration. โ
Stream"
X Link 2025-08-01T16:43Z 10.1K followers, [--] engagements
"I don't think people understand to what degree sloppy agents effect quality of llms. Take qwen-coder run it thorough different coding agents and you get a huge variance. This boils to down to [--] thing: Overengineering the agent behaviour instead of only relying on the models internal capabilities. Lobotimization is real ๐คฏ"
X Link 2025-08-01T18:17Z 10.1K followers, [----] engagements
"Hey @dagger_io team we recently released Dagger support for @vibekit_sh. A bunch of vibekit users use Dagger as their preferred sandbox for running coding agents. Would like to come in contact with someone at the team regarding our dagger package. Would be amazing if you could give feedback and contribute open source. https://github.com/superagent-ai/vibekit/tree/%40vibe-kit/sdk%400.0.56/packages/dagger https://github.com/superagent-ai/vibekit/tree/%40vibe-kit/sdk%400.0.56/packages/dagger"
X Link 2025-08-03T11:49Z 10.1K followers, [---] engagements
"@tilcrypto @elonmusk @grok i would call it something else"
X Link 2025-08-04T19:25Z 10.1K followers, [--] engagements
"@OverfitForTruth @mickeyxfriedman @elonmusk Should prob delete LinkedIn as well"
X Link 2025-08-04T19:25Z 10.1K followers, [--] engagements
"@OverfitForTruth @mickeyxfriedman @elonmusk a Mans Man"
X Link 2025-08-04T19:26Z 10.1K followers, [--] engagements
"@YTCodeAntonio @shadcn @convex_dev You are the best shipper of all time"
X Link 2025-08-04T19:35Z 10.1K followers, [---] engagements
"The simulation is real https://deepmind.google/discover/blog/genie-3-a-new-frontier-for-world-models/ https://deepmind.google/discover/blog/genie-3-a-new-frontier-for-world-models/"
X Link 2025-08-05T15:37Z 10.2K followers, [---] engagements
"The step OpenAI took was huge now just waiting for the "safest" AI in the world to do the same"
X Link 2025-08-05T17:48Z 10.2K followers, [---] engagements
"Me eating popcorn watching the AI labs this week๐ฟ"
X Link 2025-08-07T14:52Z 10.3K followers, [----] engagements
"It's to late "vibe coding" has become the default word for coding with ai. Same thing happened to "RAG" or "AGENT". Even though all these [--] works have a very specific def it's hard to take it back. "Vibe coding" does not mean any coding with AI. If you are vibing you're not looking at the code. Building something throwaway or a prototype. That is very different from using AI to write code that is reviewed tested and maintained over time. https://t.co/0rOB769AEs "Vibe coding" does not mean any coding with AI. If you are vibing you're not looking at the code. Building something throwaway or a"
X Link 2025-08-08T12:52Z 10.4K followers, [----] engagements
"So OpenAI updated their CLI Cursor just dropped theirs. Even Groq has one. I made a community drive one for Grok since XAI won't. I'm willing to hand it over to @xai for free. https://x.com/pelaseyed/status/1947283836997337199 Announcing Grok CLI ๐ฅ An open-source AI agent that brings the power of @grok directly into your terminal. Built over the weekend with a few design principles in mind: โ
No LLM frameworks โ
No lobotomizing models โ
Hackable (MIT licensed) Github link below. https://t.co/jCGK5b8VjG https://x.com/pelaseyed/status/1947283836997337199 Announcing Grok CLI ๐ฅ An open-source"
X Link 2025-08-08T13:47Z 10.3K followers, [----] engagements
"My earlier point about the word "Vibe Coding" @leerob"
X Link 2025-08-08T14:30Z 10.3K followers, [----] engagements
"Im shipping something pretty cool next week"
X Link 2025-08-09T18:34Z 10.5K followers, [----] engagements
"Would be cool if @grok imagine could generate text in images"
X Link 2025-08-10T08:50Z 10.4K followers, [----] engagements
"@TurnerNovak Rockstar taking so long that we invented and got bored of AGI before they could finish a game"
X Link 2025-08-10T13:37Z 10.4K followers, [--] engagements
"This is the real danger of LLM memory attacks. Not just data leaks but persistent control over your trusted AI relationships. memory implants go beyond the infected session to hijack any future session but its worse than just compromising your data memories are typically inserted into system instructions giving the attacker a "custom instructions" feature -- on your trusted assistant memory implants go beyond the infected session to hijack any future session but its worse than just compromising your data memories are typically inserted into system instructions giving the attacker a"
X Link 2025-08-10T14:43Z 10.4K followers, [---] engagements
"@billyuchenlin @grok Please ship CLI for Grok. The terminal interface would be a game changer for developers"
X Link 2025-08-10T17:49Z 10.4K followers, [---] engagements
"Claude Code has some interesting system prompts. The system-reminder is especially interesting. As well as passing the current git status and other system information to the system prompt"
X Link 2025-08-12T19:23Z 10.4K followers, [----] engagements
"@robj3d3 And knowing when to say no. I've found ruthless prioritization is the hardest skill to master"
X Link 2025-08-13T08:05Z 10.4K followers, [--] engagements
"@thdxr Same. Half my week is spent making unreliable MCP implementations work. Wasn't this supposed to make our lives easier"
X Link 2025-08-13T08:13Z 10.4K followers, [---] engagements
"Excited to bring VibeKit CLI support for @OpenAI codex โก npm i -g vibekit โ
Realtime redaction of API keys + sensitive data โ
Observability dashboard (runs file changes) โ
Sandboxing your local environment"
X Link 2025-08-14T13:36Z 10.5K followers, [----] engagements
"@donvito @ollama @lmstudio Same experience here. LM Studio with GGUF Q8 runs smoother than Ollama for these smaller models. Memory consumption looks reasonable too"
X Link 2025-08-15T06:42Z 10.5K followers, [---] engagements
"@donvito @ollama what's the speed when you run it"
X Link 2025-08-15T08:54Z 10.4K followers, [--] engagements
"@garyfung Why use a wrapper on AWS when you can just use AWS"
X Link 2025-08-15T15:34Z 10.4K followers, [---] engagements
"You can just use AWS ๐ I used to hate wiring CI/CD pipelines just to deploy code to AWS or GCP. Always defaulted to easy platforms like Vercel or Railway but paid the price in $$$. Now I can just vibe-code my own pipeline straight to bare metal. Faster cheaper and way more satisfying. 1/ From https://t.co/F68mGnZ4MD I used to hate wiring CI/CD pipelines just to deploy code to AWS or GCP. Always defaulted to easy platforms like Vercel or Railway but paid the price in $$$. Now I can just vibe-code my own pipeline straight to bare metal. Faster cheaper and way more satisfying. 1/ From"
X Link 2025-08-15T15:34Z 10.4K followers, [---] engagements
"@paulg Vibe coding works because it focuses on shipping instead of perfection. I've seen more success with quick messy MVPs than perfect codebases that never launch"
X Link 2025-08-16T11:09Z 10.4K followers, [---] engagements
"The vibe hole is real. I saw it. Hero dosed vibe coding. Dont come here"
X Link 2025-08-16T14:00Z 10.5K followers, [----] engagements
"https://github.com/superagent-ai/grok-cli I think @xAI is underselling Grok 4's coding abilities. They should create a tool like Claude Code to make Grok [--] shine. To get started I suggest forking Codex-CLI which is open source. This way they could get a product out of the door quickly. What do you think @elonmusk https://github.com/superagent-ai/grok-cli I think @xAI is underselling Grok 4's coding abilities. They should create a tool like Claude Code to make Grok [--] shine. To get started I suggest forking Codex-CLI which is open source. This way they could get a product out of the door"
X Link 2025-08-17T09:04Z 10.5K followers, [----] engagements
"You can now use Cursor CLI and Opencode without wrecking anything โ
Stops destructive commands from trashing your local machine โ
Redacts secrets and API keys in real time โ
Lets you actually see whats going on npm i -g vibekit"
X Link 2025-08-19T13:33Z 10.5K followers, [---] engagements
"Gemma [--] (270M) looks like a big step for compact AI chips. With lower latency and training time/cost you can now basically train models in realtime. Imagine deploying agents to the edge"
X Link 2025-08-19T18:07Z 10.5K followers, [---] engagements
"@BrendanFalk Looks cool what sandbox are you using"
X Link 2025-08-19T19:59Z 10.5K followers, [--] engagements
"@BrendanFalk was wondering if you had checked out E2B Daytona etc. (dedicated sandboxes for generating code w/ agents)"
X Link 2025-08-19T20:06Z 10.4K followers, [--] engagements
"Funny how the feature everyone praises in VibeKit the one that gets all the attention was just a last-minute patch I threw together to fix a sandbox escape no one reported for weeks"
X Link 2025-08-20T05:49Z 10.5K followers, [---] engagements
"@AlexReibman @warpdotdev I use warp everyday but I'm not sure about terminal bench. I vibe coded Grok CLI in a weekend and it scored better than Gemini CLI while being obviously worse ๐"
X Link 2025-08-20T06:02Z 10.5K followers, [---] engagements
"This simple prompt injection got read/write access to 1M repositories. Simple static analysis tools won't spot it. Your LLM observability layer that promises "security" might flag it but by then the damage is done. In order truly secure model inputs/outputs and actions you need to guard against injections prior to them ever happening. This is a difficult task especially due to the sheer amount of code that agents write making it super hard for humans to review . There is a better way"
X Link 2025-08-21T14:00Z 10.5K followers, [----] engagements
"While testing prompt injection techniques I found Cursor runs shell commands straight from files ๐คฏ I was experimenting with different injection techniques for a model dataset and came across something concerning. If a file contains instructions like run this shell command Cursor doesnt stop to ask or warn you. It just runs it. Directly on your local machine. That means if you: 1) Open a malicious repo 2) Ask to summarize or inspect a file Cursor could end up executing arbitrary commands including things like exfiltrating environment variables or installing malware. To be clear: - Ive already"
X Link 2025-08-22T11:44Z 10.5K followers, [----] engagements
"@nathanhamiel This was not expected While testing prompt injection techniques I found Cursor runs shell commands straight from files ๐คฏ I was experimenting with different injection techniques for a model dataset and came across something concerning. If a file contains instructions like run this shell https://t.co/LukevGhyBM While testing prompt injection techniques I found Cursor runs shell commands straight from files ๐คฏ I was experimenting with different injection techniques for a model dataset and came across something concerning. If a file contains instructions like run this shell"
X Link 2025-08-22T11:50Z 10.5K followers, [--] engagements
"@LeonDerczynski After this I managed to manipulate the task list by injecting in comments. Im pretty sure I can do more or less anything from here on out. The system susceptible to backdoors etc. I posted to the Cursor Reddit but just got hate back ๐"
X Link 2025-08-23T14:06Z 10.5K followers, [--] engagements
"There is no point for the CEO to make more the state taxes the crap out of everything over a certain threshold. This does not take into account "asset hording" by high-income people (stocks real-estate etc.). Norway is a country of [---] Million people highly subsidized by Oil. In Norway CEOs make 11x more than their average worker. In the United States CEOs make 351x more than their average worker. In Norway CEOs make 11x more than their average worker. In the United States CEOs make 351x more than their average worker"
X Link 2025-08-25T06:49Z 10.5K followers, [----] engagements
"@leoubbiali Going to Rome in Oct Stadio Olimpico is calling"
X Link 2025-08-25T06:58Z 10.5K followers, [--] engagements
"@Wrestlephobia3 You can compare the both but comparing on income but majority of wealth is in assets is my point (stocks real-estate etc. etc.) I doubt norway is much better than the US"
X Link 2025-08-25T07:36Z 10.5K followers, [--] engagements
"It would be cool if I could use GPT models inside of Claude code"
X Link 2025-08-26T07:47Z 10.5K followers, [----] engagements
"@wunderwuzzi23 Yes some others have the exact same issues exposing MCP servers and having tokens hardcoded into the sandbox not sure why one needs to hardcode these tokens at all"
X Link 2025-08-26T11:53Z 10.5K followers, [--] engagements
"@bindureddy Gemma [--] 270m is a beat imo"
X Link 2025-08-28T09:20Z 10.5K followers, [---] engagements
"Garry's point is that everything that is publicly available to a human should be publicly available to an Agent without having to go through an identification layer. Also Garry + Cloudflare had some exchanges about CF blocking AI bots per default (which later turned out to be a bug)"
X Link 2025-08-29T06:40Z 10.5K followers, [----] engagements
"@gdb you should've shipped it as a terminal ui imo"
X Link 2025-08-29T09:29Z 10.5K followers, [---] engagements
"@keithwhor You dont want your lawyer being paid by the counter party. Also its priced in in the series a"
X Link 2025-08-29T19:50Z 10.5K followers, [--] engagements
"@Yuchenj_UW Soon"
X Link 2025-08-30T06:52Z 10.5K followers, [----] engagements
"@amasad Big RN fan I had to go around town and convince people it was actually native back then ๐คฃ They were burned by Senja HTML5 etc. etc. I absolutely loved porting native SDKs. Made the official Twilio RN CallKit SDK"
X Link 2025-09-01T20:23Z 10.5K followers, [---] engagements
"@thisismahmoud_ @Railway This was my bad the image amd64/linux was breaking so Railway couldn't find the repo"
X Link 2025-09-01T21:05Z 10.5K followers, [--] engagements
"@JustJake @thisismahmoud_ @Railway It would be amazing if I could attach a GPU to a service (I guess you've heard that before). One of my services runs a small edge model (Gemma [--] 270M) it can run on CPU but would go brrrr if the service had some more juice"
X Link 2025-09-01T21:23Z 10.5K followers, [--] engagements
"@corbtt @OpenPipeAI @CoreWeave Congrats"
X Link 2025-09-03T20:47Z 10.5K followers, [--] engagements
"Prompt injection is doing to AI what XSS once did to the web ๐ฉ A new paper on Cybersecurity AI: Hacking the AI Hackers via Prompt Injection caught my attention. It shows how large language models can be manipulated through hidden instructions in the same way websites were once vulnerable to cross-site scripting. What stood out is not just that the attacks work but how structural the problem is. Prompt injection is not a bug that can be patched once. It is a class of vulnerability that emerges whenever models interact with untrusted inputs. History repeats itself. Web security only matured"
X Link 2025-09-04T13:51Z 10.5K followers, [----] engagements
"@dok2001 @Cloudflare The internet was built on cooperation not corporate silos. Keep that spirit alive"
X Link 2025-09-05T06:15Z 10.5K followers, [---] engagements
"The riskiest AI in your stack might be a chat widget you barely remember approving. The Salesloft Drift compromise let attackers steal OAuth and refresh tokens through Salesforce connectors and pivot into customer data at multiple firms including Zscaler"
X Link 2025-09-05T14:00Z 10.5K followers, [---] engagements
"@ghostwriternr @mattzcarey @zeror404 @Cloudflare @vibekit_sh Yup @vibekit_sh enables you to run cc codex etc with cloudflare sandboxes. https://docs.vibekit.sh/supported-sandboxes/cloudflare https://docs.vibekit.sh/supported-sandboxes/cloudflare"
X Link 2025-09-08T15:37Z 10.5K followers, [--] engagements
"@prem_k @threepointone That would indeed be sick. We use Gemma as base model in one of our products and the cost/intelligence ratio is quite significant. Im bullish on these smaller edge models (personally think the edge is the future)"
X Link 2025-09-08T16:28Z 10.5K followers, [--] engagements
"As these coding agents gradually get introduced into the enterprise all it takes is this is literally all it takes to get access to their systems. 1) Figure out which coding agent company X is using. 2) Pinpoint employees at that specific company. 3) Have that employee summarize some phony url with their agent and inject instructions. I would put money on this already happening"
X Link 2025-09-08T16:42Z 10.5K followers, [----] engagements
"This is just the tip of the ice berg. you ain't seen nothing yet. Picture this: Someone injects malicious tokens into a super popular package website or anything else that will get consumed by an agent (web research tool of claude code chatgpt or whatever). The attacker can reach tens of thousands of users"
X Link 2025-09-08T17:48Z 10.5K followers, [--] engagements
"Ninja-lm ๐ฅท is trained on a carefully curated dataset of prompt injections and backdoors making any Agent highly resilient against malicious attacks. Its optimized to run on CPU making it ideal for real-time threat detection at the edge"
X Link 2025-09-09T13:03Z 10.5K followers, [---] engagements
"You can download it on @huggingface and run it on their inference endpoints with one click. https://huggingface.co/superagent-ai/ninja-lm-270m-gguf https://huggingface.co/superagent-ai/ninja-lm-270m-gguf"
X Link 2025-09-09T13:03Z 10.5K followers, [---] engagements
"A majority of hacks occur when an agent does tool calls to fetch content from the web. Webpages containing hidden injections etc. like this one. Would be cool to run a security scan on @ExaAILabs indexes @WilliamBryk wdyt"
X Link 2025-09-11T15:00Z 10.5K followers, [---] engagements
"Replit is the by far the best vibe coding tool out there. It gives non-developers the perfect ecosystem to build on auth db production deployments etc etc. With Agent [--] they take this to the next level with they testing agent something. The cursor claude code Lovable etc misses. This new modality will prove to be very important. Yesterday asked it to build a YouTube clone took [--] minutes uploads auth search explore worked out of the box. My two take-aways are 1) Current LLMs are very underutilized 2) Building these systems of agents with different scopes/modalities can 10x the output you get"
X Link 2025-09-11T15:25Z 10.5K followers, [---] engagements
"We just shipped a bunch of updates to Grok CLI ๐ฅ - Faster and more accurate diffs - Run headless - Bunch of bug fixes npm i -g vibe-kit/grok-cli"
X Link 2025-09-12T14:30Z 10.5K followers, [---] engagements
"@tejasybhakta Yep yolo is inevitable with the current UX"
X Link 2025-09-12T20:13Z 10.5K followers, [--] engagements
"@AIadventurer Makes sense the current UX won't cut it"
X Link 2025-09-13T07:42Z 10.5K followers, [--] engagements
"Replit is the only vibe coding tool that figured out not set env variables on the client side. Many such cases in the other ones"
X Link 2025-09-13T07:52Z 10.5K followers, [---] engagements
"@rauchg Bring the vibes to data sounds compelling especially compared to a bunch of lists/charts or other general UX"
X Link 2025-09-15T19:48Z 10.5K followers, [---] engagements
"@elonmusk The Swedish Pension Funds (state owned investment body that gambles with peoples money) sold TSLA stocks has had +100% increase since. That's what you get when you let the government act as day traders"
X Link 2025-09-16T06:15Z 10.5K followers, [---] engagements
"Sweden has produced some cracked founders per capita: Databricks Cursor Modal Legora Lovable Sana"
X Link 2025-09-17T17:00Z 10.5K followers, [----] engagements
"@thdxr The deep integration with with the terminal. But agree the instructions would be a problem e.g something like Manus"
X Link 2025-09-18T10:24Z 10.5K followers, [----] engagements
"@thdxr Do you have any docs for opentui Would like to implement in a project"
X Link 2025-09-19T07:35Z 10.5K followers, [---] engagements
"@thdxr Ara is the GOAT of shitting on Open Source initiatives. How wrote a book about how I suck deceptive etc. etc. and sent it to me in a DM ๐คฃ๐คฃ"
X Link 2025-09-25T06:36Z 10.5K followers, [---] engagements
"@whoiskatrin @Cloudflare congrats and keep shipping ๐ซก"
X Link 2025-09-25T14:26Z 10.5K followers, [---] engagements
"@deedydas I think we need WorkBench eval AI Agents that complete tasks on Upwork"
X Link 2025-09-26T05:37Z 10.5K followers, [---] engagements
"The only way is to become one with the AI. https://nmn.gl/blog/vibe-coding-gambling https://nmn.gl/blog/vibe-coding-gambling"
X Link 2025-09-26T05:46Z 10.5K followers, [---] engagements
"@Benioff @SlackHQ @agentforce @salesforce Lets make Agentforce secure AI makes shipping 10x faster thats what most of us care about atm. This could have been mitigated with one line of code. await guard(prompt) AI makes shipping 10x faster thats what most of us care about atm. This could have been mitigated with one line of code. await guard(prompt)"
X Link 2025-09-26T21:05Z 10.5K followers, [---] engagements
"Is there a nicer way to show token usage cost in the billing page @polar_sh @birk"
X Link 2025-09-29T12:10Z 10.5K followers, [---] engagements
"So the whole chunk split embed retrieve rerank was just a bunch of BS"
X Link 2025-09-29T20:24Z 10.5K followers, [---] engagements
"Grok CLI made its way into @TaskmasterAI ๐ฅ @EyalToledano is building some awesome stuff and I'm pretty sure it will be huge [--] grok cli support (s/o @pelaseyed) generate/update tasks w/ full codebase context latest @grok models (grok-2 grok-3 grok-4) tm models --set-main grok-beta for AI ops grok cli was built by @pelaseyed check out his repo https://t.co/5eyvLlswjr works awesome tbh https://t.co/a72KdFI6N6 [--] grok cli support (s/o @pelaseyed) generate/update tasks w/ full codebase context latest @grok models (grok-2 grok-3 grok-4) tm models --set-main grok-beta for AI ops grok cli was built"
X Link 2025-10-01T18:11Z 10.5K followers, [---] engagements
"We shipped one of the most asked for features: Realtime Redaction ๐คซโก Make your AI Agents compliant with AI Act SOC-2 GDPR etc. with a couple of lines of code"
X Link 2025-10-02T12:58Z 10.5K followers, [--] engagements
"Checkout or docs for more details: https://docs.superagent.sh/typescript-sdk https://docs.superagent.sh/typescript-sdk"
X Link 2025-10-02T12:58Z 10.5K followers, [--] engagements
"We shipped one of the most asked for features: Runtime Redaction. Automatically remove sensitive data from prompts responses and tool calls. โ
PII/PHI โ
API Keys and Secrets โ
Credit cards IP addresses and much more"
X Link 2025-10-02T14:30Z 10.5K followers, [---] engagements
"Compliance and security now made simple for @mastra_ai users Ive followed Mastras journey closely and its clear theyre building one of the most ambitious and well-architected AI Clouds out there. Here's how to secure your Mastra Agents in [--] min:"
X Link 2025-10-03T12:57Z 10.5K followers, [----] engagements
"@red___square @mastra_ai 0.2cents per millon tokens 100-400ms based on the size of the payload. Redaction is more or less instant"
X Link 2025-10-03T13:11Z 10.5K followers, [--] engagements
"@red___square @mastra_ai Also you can host the models yourself they are open source"
X Link 2025-10-03T13:20Z 10.5K followers, [--] engagements
"@jaredpalmer @Microsoft @github @code @Azure Congrats"
X Link 2025-10-13T14:45Z 10.5K followers, [---] engagements
"@aditabrm @reductoai @a16z Big congrats"
X Link 2025-10-14T16:20Z 10.5K followers, [---] engagements
"Gifted progress bar for React Native #reactnative https://www.npmjs.com/package/react-native-gifted-progress https://www.npmjs.com/package/react-native-gifted-progress"
X Link 2017-02-06T10:04Z [----] followers, [--] engagements
Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing
/creator/twitter::pelaseyed