@PawelHuryn Paweล Huryn
Paweล Huryn posts on X about ai, agentic, openclaw, how to the most. They currently have [------] followers and [---] posts still getting attention that total [-----] engagements in the last [--] hours.
Engagements: [-----] #
- [--] Week [-------] +136%
- [--] Month [-------] +5,323%
- [--] Months [---------] -58%
- [--] Year [---------] +37%
Mentions: [--] #
- [--] Month [--] -86%
- [--] Months [---] +5.30%
- [--] Year [---] +93%
Followers: [------] #
- [--] Week [------] +4%
- [--] Month [------] +15%
- [--] Months [------] +35%
- [--] Year [------] +71%
CreatorRank: [---------] #
Social Influence
Social category influence technology brands 31% finance 7% stocks 6% social networks 4% events 1%
Social topic influence ai 37%, agentic 8%, openclaw #1391, how to 7%, in the 6%, open ai #2115, llm 5%, tools 5%, anthropic #2658, stocks 4%
Top accounts mentioned or mentioned by @busterr97641 @hamelhusain @mathemba @clerk @karakhanyans @cryptotradercb8 @tendies @thelyons @shreya @mariusorani @logikaagency @sooyoon_eth @joshualelon @n8nio @openrouterai @saendev @web3youth @currenticai @alk000 @openai
Top assets mentioned Alphabet Inc Class A (GOOGL) Microsoft Corp. (MSFT) Cloudflare, Inc. (NET)
Top Social Posts
Top posts by engagements in the last [--] hours
"Enjoy this [--]. Follow me @PawelHuryn + ๐ [--]. Share this thread with friends I appreciate it https://x.com/PawelHuryn/status/2021247927176368637s=20 RIP OpenClaw. Introducing Agent One: Autonomy with Security. Built in [--] days with Opus [---] and n8n. A short demo: ๐งต https://t.co/itQrvFvwyA https://x.com/PawelHuryn/status/2021247927176368637s=20 RIP OpenClaw. Introducing Agent One: Autonomy with Security. Built in [--] days with Opus [---] and n8n. A short demo: ๐งต https://t.co/itQrvFvwyA"
X Link 2026-02-10T15:45Z 36.6K followers, [----] engagements
"RT @PawelHuryn: Anthropic was founded by people who left OpenAI over safety concerns. Now Anthropic's safety lead is leaving Anthropic ove"
X Link 2026-02-12T07:39Z 36.6K followers, [---] engagements
"RIP OpenClaw. How to use Claude Opus [---] + n8n to create a secure autonomous agent available on all your devices: Step 1: Install Desktop Commander (Docker) Step 2: Configure permissions (mounted folders) Step 3: Create a secure connection Works today: - Run mcp-proxy - Set up a Cloudflare tunnel - Use a custom domain (from $5/year) An alternative coming soon: Use Desktop Commander Remote (free Beta) Step 4: Create an n8n agent with Opus 4.6: - n8n VPS ($4.99/mo) - Add a Telegram trigger - Add Desktop Commander MCP - Plug integrations (Gmail Drive Notion Stripe) - Add memory subagents and the"
X Link 2026-02-08T13:20Z 36.6K followers, 211.8K engagements
"RIP OpenClaw. Introducing Agent One: Autonomy with Security. Built in [--] days with Opus [---] and n8n. A short demo: ๐งต"
X Link 2026-02-10T15:40Z 36.6K followers, 72.9K engagements
"It's here: https://x.com/PawelHuryn/status/2021247927176368637s=20 RIP OpenClaw. Introducing Agent One: Autonomy with Security. Built in [--] days with Opus [---] and n8n. A short demo: ๐งต https://t.co/itQrvFvwyA https://x.com/PawelHuryn/status/2021247927176368637s=20 RIP OpenClaw. Introducing Agent One: Autonomy with Security. Built in [--] days with Opus [---] and n8n. A short demo: ๐งต https://t.co/itQrvFvwyA"
X Link 2026-02-10T15:47Z 36.6K followers, [---] engagements
"Anthropic was founded by people who left OpenAI over safety concerns. Now Anthropic's safety lead is leaving Anthropic over safety concerns. Same week: Claude Cowork wiped $285 billion off SaaS stocks. Anthropic is closing a $20 billion raise at $350 billion. His letter: the organization "constantly faces pressures to set aside what matters most." That pressure has a name. A $350 billion roadmap that won't slow down. Today is my last day at Anthropic. I resigned. Here is the letter I shared with my colleagues explaining my decision. https://t.co/Qe4QyAFmxL Today is my last day at Anthropic. I"
X Link 2026-02-11T10:55Z 36.6K followers, 203K engagements
"OpenClaw has 186K GitHub stars and 1.5M compromised API keys. I needed a secure alternative. So I built it with n8n and Claude Opus [---]. It can already: - Reply to your Telegram messages - Access selected folders from your laptop - Access Gmail Drive Notion Linear etc. - Install new local tools in a sandbox - Run autonomously for hours - Create multiple subagents - Learn from experience - Wake up regularly But unlike OpenClaw it: - Can't access your API keys - Can't modify its environment - Can't access folders you haven't shared - Can't access tools you haven't approved - Must get your"
X Link 2026-02-12T13:46Z 36.6K followers, 50.5K engagements
"A feature everyone missed. One-click observability for @n8n_io Moltbolt custom apps or even Claude Code (yes it can work with @OpenRouterAI). What catched my attention: sending your traces not just to common providers but also to your webhooks - custom data viewer. ๐ Announcing "Broadcast": fan-out LLM observability Broadcast is a simple way to automatically send traces from your OpenRouter requests to external platforms with no additional code. https://t.co/vxVv86Ekzt ๐ Announcing "Broadcast": fan-out LLM observability Broadcast is a simple way to automatically send traces from your"
X Link 2026-01-28T19:16Z 35.6K followers, [----] engagements
"Moonshot AI just released Kimi K2.5. Open-source. Beats GPT-5.2 on most benchmarks. Matches Claude Opus [---] on coding. I've been testing it with VS Code. The coding output is real. This isn't hype. Here's why this matters: frontier AI can be built cheap. DeepSeek cost under $6M to train. These teams charge pennies for API access and still make margins. Every company charging premium rates for AI features is now competing against near-free. The open-source strategy is calculated. Chinese AI companies learned that Western enterprises won't touch their APIs. Data security concerns are too high."
X Link 2026-01-29T06:12Z 35.6K followers, [----] engagements
"OpenClaw security vulnerability that blew my mind: - It claims it can't ignore hard guardrails (e.g. exec approval). - But when asked to modify them it does it without hesitating for a moment. Recommended: [--]. Never install OpenClaw on your main machine. This needs to be an isolated environment. [--]. Never share your personal tokens with OpenClaw. It should use its dedicated accounts (e.g. Gmail) and API keys. --- OpenClaw is worth your attention not because it's production-ready (it's not) but because it's the cheapest way for PMs and engineers to build intuition for AI that's coming in 2026."
X Link 2026-02-03T00:21Z 36.3K followers, 15.5K engagements
"RE: "Anthropic just wiped $285B off the stock market with a GitHub plugin that automates NDA triage"* My BS score: 6/10 I'm a PM not an analyst. After a brief research here's what the market is completely ignoring and what matters from an AI product strategy perspective: "AI can draft a contract summary" "AI replaces proprietary legal research infrastructure." The moat isn't just the data - it's the trust. Lawyers cite Westlaw and LexisNexis in court filings. Courts accept them as authoritative. No lawyer is citing "Claude said so" in front of a judge. Raw case law is public. But decades of"
X Link 2026-02-05T11:16Z 36.4K followers, [----] engagements
"I never thought you can be excited about the pricing update email. Clerk just sent me: - 5x limits - Free MFA vs. $100/mo add-on - Free satellite domains vs. $100/mo - Click here to opt in A new bill: $225/mo (planned) $20/mo (new) You rock @clerk ๐๐ Who's next"
X Link 2026-02-06T00:39Z 35.9K followers, [---] engagements
"For anyone serious about their data a detailed guide on installing OpenClaw. If you haven't read it consider rotating all your API keys first. malware found in the top downloaded skill on clawhub and so it begins https://t.co/VY4EeWExro malware found in the top downloaded skill on clawhub and so it begins https://t.co/VY4EeWExro"
X Link 2026-02-07T08:45Z 35.4K followers, [---] engagements
"@karakhanyanS Do you install OpenClaw on your physical machine ๐ฒ"
X Link 2026-02-08T20:08Z 35.8K followers, [---] engagements
"@saen_dev OpenRouter + limits on dedicated keys You can also consider Civic Nexus MCP gateway for governance"
X Link 2026-02-08T20:10Z 36.5K followers, [----] engagements
"@CryptotraderCB8 Need to test that. True Opus east credits fast"
X Link 2026-02-08T20:10Z 35.8K followers, [---] engagements
"@tendies Will be automated. The point is protecting your data and credentials. Some things can't be measured in clicks"
X Link 2026-02-08T20:18Z 35.8K followers, [---] engagements
"@Web3__Youth Interesting. Yes if API allows it. Need to test that"
X Link 2026-02-08T20:18Z 35.7K followers, [---] engagements
"@TheLyons Did you download the Twitter skill ;)"
X Link 2026-02-08T20:21Z 35.9K followers, [---] engagements
"@CurrenticAI I'd define it like this: anyone can spend 2-3 days to build a more secure system with similar capabilities - even in n8n"
X Link 2026-02-08T20:23Z 36.2K followers, [----] engagements
"@al_k_000 Not really. You decide what requires your approval. It has a memory and can learn new skills. It can also install new tools in the sandbox. But yes you need to configure every integration without revealing API keys to the agent"
X Link 2026-02-08T20:27Z 36.4K followers, [---] engagements
"It can: - Reply to your Telegram or Slack messages - Access selected folders from your laptop - Access Gmail Drive Notion Linear etc. - Install new local tools in a sandbox - Run autonomously for hours - Create multiple subagents - Learn from experience - Wake up regularly"
X Link 2026-02-10T15:40Z 36.5K followers, [----] engagements
"I wanted an autonomous agent available on all my devices. But I didn't want: - [-----] emails and 1.5M API keys exposed - The top-downloaded community skill Malware Agent One: - Can't access your API keys - Can't modify its environment - Can't access folders you haven't shared - Can't access tools you haven't approved - Must get your confirmation e.g. when sending emails These arent prompt instructions. Theyre hard architectural boundaries Docker isolation mounted folder permissions n8ns tool approval system. https://twitter.com/i/web/status/2021247935263146379"
X Link 2026-02-10T15:40Z 36.5K followers, [----] engagements
"The VPS on Hostinger hosts n8n and sandbox containers. The sandbox is available only from n8n - this is where VPS Executor agent can manage files install new tools and execute scripts"
X Link 2026-02-10T15:40Z 36.5K followers, [----] engagements
"Data Tables in n8n store both memories and sessions no external database no vector store no infrastructure. Just rows in a table. Memory is the long-term knowledge. Two types: - Manager memory: user preferences facts corrections project context - Executor environment memory: what tools are installed whats broken workarounds discovered"
X Link 2026-02-10T15:40Z 36.2K followers, [----] engagements
"2. Implement hard guardrails vs. soft suggestions Ask me before sending an email as a prompt instruction = a suggestion. The model ignores it under pressure. Wrapping it in an orchestrated gateway = a hard guardrail. The model cant skip it:"
X Link 2026-02-10T15:40Z 36.2K followers, [----] engagements
"3. Minimal contract for multi-agent coordination The ManagerExecutor interface is [--] fields: context goal constraints. I tried adding more: - executor_tasks (array of sub-tasks) - internal_notes (reasoning for next iteration) - env_context (what tools are installed). Each addition made the system worse. The model spent more tokens managing the contract than doing the work. Keep contracts minimal. https://twitter.com/i/web/status/2021247965998768159 https://twitter.com/i/web/status/2021247965998768159"
X Link 2026-02-10T15:40Z 36.2K followers, [----] engagements
"Extensions easy to add: - Heartbeat via Cron Agent One can wake up periodically to analyze recent sessions check emails review calendar surface whats important. Proactive not just reactive. It would be just a simple scheduled prompt. - Civic Nexus governance a framework for managing agent permissions audit trails and escalation policies as the system grows. Looks promising. - Supermemory integration when the simple Data Table memory isnt enough for semantic recall. - WhatsApp as an additional surface Telegram works but WhatsApp reaches more people"
X Link 2026-02-10T15:40Z 36.2K followers, [----] engagements
"The point isn't that Agent One is production-ready. It's that the architecture is. Docker isolation mounted folder permissions tool approval structured contracts none of this is new technology. It's just discipline. OpenClaw proved people want personal AI agents. It also proved that just trust the prompt isnt a security model. The line between an autonomous agent and a dangerous one isnt a prompt. Its a design choice. https://twitter.com/i/web/status/2021247981576450157 https://twitter.com/i/web/status/2021247981576450157"
X Link 2026-02-10T15:40Z 36.2K followers, [----] engagements
"Codex may be a better coding model. So why does Opus win on usage Because the best technology has never won a platform war. The best integration into daily work has. Google beat AltaVista not by better search results by becoming the homepage. Slack beat email not by better messaging by becoming where work happened. Opus handles files browses the web builds agents runs workflows. Codex writes better code in a sandbox. One is a tool. The other is infrastructure. codex is by far a better coding model than opus - anyone who knows anything understands this but the whole industry should reflect on"
X Link 2026-02-12T07:07Z 36.2K followers, [---] engagements
"Nick Huber's [--] entrepreneurs didn't mention AI once. [--] days ago a set of Markdown plugins wiped $285 billion from software stocks. Not a new model. Config files. Karpathy says he's 10x more productive. Developers stopped writing code by hand. A reasoning test designed to be unsolvable went from 5% to 84% in [--] months (Gemini [--] Deep Think on ARC-AGI-2). We've never been such close to singularity. His [--] entrepreneurs not talking about AI tells you who was in the room. Not what's happening outside it. The AI hype on X is such a bubble. It is starting to feel like the pandemic panic. I just"
X Link 2026-02-14T17:53Z 36.4K followers, [--] engagements
"Nick Huber's [--] entrepreneurs didn't mention AI once. [--] days ago a set of Markdown plugins wiped $285 billion from software stocks. Not a new model. Config files. Karpathy says he could be 10x more productive. Developers stopped writing code by hand. A reasoning test designed to be unsolvable went from 5% to 84% in [--] months (Gemini [--] Deep Think on ARC-AGI-2). Nick's [--] entrepreneurs not talking about AI tells you who was in the room. Not what's happening outside it. The AI hype on X is such a bubble. It is starting to feel like the pandemic panic. I just hung out with [--] successful"
X Link 2026-02-14T17:56Z 36.4K followers, [---] engagements
"Nick Huber's [--] entrepreneurs didn't mention AI once. [--] days ago a set of Markdown plugins wiped $285 billion from software stocks. Not a new model. Config files. Karpathy says he could be 10x more productive. A reasoning test designed to be unsolvable went from 5% to 84% in [--] months (Gemini [--] Deep Think on ARC-AGI-2). Nick's [--] entrepreneurs not talking about AI tells you who was in the room. Not what's happening outside it. The AI hype on X is such a bubble. It is starting to feel like the pandemic panic. I just hung out with [--] successful entrepreneurs for [--] days and didnt hear the word"
X Link 2026-02-14T18:07Z 36.4K followers, [---] engagements
"Sessions are short-term state for multi-step tasks. Original request plan assumptions and a log. When the Manager loops with fresh context the session is all it gets"
X Link 2026-02-10T15:40Z 36.5K followers, [----] engagements
"For complex multi-step tasks Agent One uses my variant of the Ralph Wiggum loop. When the Manager outputs Continue the orchestrator resets its context completely. The next iteration has no memory of what just happened except the session. The session is everything. Original request goal plan with statuses assumptions log. The orchestrator loads it and injects it as the Managers context. https://twitter.com/i/web/status/2021247956209262961 https://twitter.com/i/web/status/2021247956209262961"
X Link 2026-02-10T15:40Z 36.5K followers, [----] engagements
"That's a wrap. Thanks for reading You can read the full guide and download all templates and setup instructions from my newsletter (premium subscribers or trial): https://www.productcompass.pm/p/secure-ai-agent-n8n-openclaw-alternative https://www.productcompass.pm/p/secure-ai-agent-n8n-openclaw-alternative"
X Link 2026-02-10T15:40Z 36.5K followers, [----] engagements
"OpenAI is shipping AI pens and social features. Then they acqui-hired the guy who built the fastest-growing GitHub repo in history on $10K-$20K a month of his own money. A company with thousands of engineers and $13 billion in compute shouldn't need an acqui-hire to solve agents. But when you're shipping everything at once nothing gets finished. Steinberger proved focus beats scale. OpenAI is betting they can buy focus instead of practicing it. Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of"
X Link 2026-02-16T01:23Z 36.5K followers, [---] engagements
"DeepSeek-V3.2 is the worst "frontier" model I've recently tested with AI agents. Depending on the test only Kimi-k2 performs worse. None of them is able to follow simple instructions: output a valid chess move without adding anything else to the output. Meanwhile GPT-5.1 and Gemini-3 Pro consistently produce long games without breaking a strict SAN format. In production this is the same skill as returning clean JSON tool calls or structured outputs without extra text. Best agentic models confirmed now and by my previous experience: #1 GPT-5.1 by @OpenAI #2 Gemini [--] Pro by @Google #3 GPT-4.1"
X Link 2025-12-03T19:02Z 30K followers, [---] engagements
"I ran a simple but brutal test: top LLMs playing chess where one invalid move = game over. The results were surprising. And they say a lot about how these models behave as AI agents. - GPT-5.1 and Gemini [--] were the most reliable - DeepSeek-V3.2 and Kimi-K2 failed basic instructions - Models played better without seeing the board ๐คฏ Full data and analysis below: ๐ [--]. BEST LLMs FOR AI AGENTS Chess obviously depends on training data but two parts of this test map directly to real agent behavior: Output discipline GPT-5.1 and Gemini-3 Pro were the only models that consistently produced long"
X Link 2025-12-03T20:39Z 30K followers, [---] engagements
"I ran a simple but brutal test: top LLMs playing chess where one invalid move = game over. The results were surprising. And they say a lot about how these models behave as AI agents. - GPT-5.1 and Gemini [--] were the most reliable - DeepSeek-V3.2 and Kimi-K2 failed basic instructions - Models played better without seeing the board ๐คฏ(more steps before an invalid move) Full data and analysis below: ๐ [--]. BEST LLMs FOR AI AGENTS Chess obviously depends on training data but two parts of this test map directly to real agent behavior: Output discipline GPT-5.1 and Gemini-3 Pro were the only models"
X Link 2025-12-03T21:11Z 30.1K followers, [----] engagements
"OpenAI has announced a red alert. Everything is paused. The focus is making ChatGPT better. Cool. But please dont call this strategy. The difference between respecting strategy and extinguishing fires is making bets on what comes next not reacting to crises. Everyone can pull their hand back when it burns. Strategy is sometimes holding the line despite the pain"
X Link 2025-12-04T23:59Z 30K followers, [---] engagements
"Lovable is my favorite no-code tool for PMs. But theres always been one blocker: You need a $50$200/month to build something useful. So I kept searching for something I could confidently recommend to my readers. And I found it"
X Link 2025-12-07T19:46Z 30.3K followers, 25.5K engagements
"Do this instead: log traces read [---] diverse ones open code cluster into failure modes repeat until you hit saturation"
X Link 2025-12-18T23:52Z 30.4K followers, [--] engagements
". Trap: "Generate [---] test queries" with no structure. Why it fails: you get happy-path coverage and miss the real breakpoints"
X Link 2025-12-18T23:52Z 30.4K followers, [--] engagements
"If the failure is critical use guardrails (sync): - Fix it live - Or block the response ๐ If a failure is critical we use guardrails. For everything else we build async evaluators to improve in quick cycles. This shows an experimentation mindset"
X Link 2025-12-19T10:46Z 30.5K followers, [--] engagements
"3. Continuous Improvement Quality isnt a one-time feature. Its a loop. You dont try to predict failures. You trace them then decide how to respond: - Ship Trace (surface real failure modes) - Trace Decide (cluster fix eval guardrail) - Decide Improve (close the loop)"
X Link 2025-12-19T10:46Z 30.5K followers, [--] engagements
"This is how modern AI teams work. No traditional PM speaks like this. AI PMs must"
X Link 2025-12-19T10:46Z 30.4K followers, [--] engagements
"How to answer How do you ensure quality Don't just say "we test it." Say: I start with error analysis to uncover failure modes. I track them with evals so we improve continuously. I isolate critical failures and wrap them in guardrails. Youll sound like someone who has actually built AI products"
X Link 2025-12-19T10:46Z 30.4K followers, [---] engagements
"Q3: When do I get it They are asking:When do we learn and when can we change course Roadmaps dont fail because timelines slip. They fail because leaders stop believing them"
X Link 2025-12-21T11:22Z 30.5K followers, [--] engagements
"In AI products quality is probabilistic. Drift is real. Integrations surprise you. Example answer: First signal in [--] weeks from a 5% pilot credible read in [--] weeks after expansion scale-or-stop by week [--]. Here's how well validate it (.)""
X Link 2025-12-21T11:22Z 30.5K followers, [--] engagements
"Your most engaged users might be killing your AI margins. In SaaS power users are a blessing. They drive retention referrals expansion. In AI power users drive variance. And variance is where your economics break"
X Link 2025-12-30T18:00Z 30.8K followers, [---] engagements
"In SaaS growth tends to smooth costs. In AI growth often amplifies them. Two customers on the same plan. One costs 10x more to serve. Not because they abuse it. Because they're engaged - the thing you optimized for in SaaS"
X Link 2025-12-30T18:00Z 30.8K followers, [--] engagements
"3. Averages lie variance kills Your system doesn't break on average cost per request. It breaks on p95 behavior: long context retries spikes edge cases that force bigger models. The uncomfortable implication: AI pricing must be designed for worst-case behavior not typical behavior"
X Link 2025-12-30T18:00Z 30.8K followers, [--] engagements
"Heres a framing most teams miss: Pricing in AI isn't just how you charge money. It's one of the strongest control mechanisms you have. Pricing teaches users what "good usage" looks like. When to use heavy reasoning how much context to include when to batch vs iterate when retries are worth it. https://twitter.com/i/web/status/2006062848791736747 https://twitter.com/i/web/status/2006062848791736747"
X Link 2025-12-30T18:00Z 30.8K followers, [--] engagements
"The most common mistake is treating pricing as something to optimize later. In AI pricing is system design not a go-to-market tweak. It decides who absorbs variance and which behaviors your system must support under real user pressure"
X Link 2025-12-30T18:00Z 30.8K followers, [--] engagements
"If you don't design pricing with the same rigor as architecture the system will expose that weakness at scale. Not immediately. Not loudly. But inevitably"
X Link 2025-12-30T18:00Z 30.8K followers, [--] engagements
"Miqdad Jaffer (Product Lead at OpenAI) goes deeper on this (no paywall parts): - Why AI pricing is different from SaaS - The 7-layer AI cost stack - The [--] pricing models that survive real usage - The rule you learn in postmortems You can read it https://www.productcompass.pm/p/ai-product-pricing https://www.productcompass.pm/p/ai-product-pricing"
X Link 2025-12-30T18:00Z 30.8K followers, [---] engagements
"Want to understand how AI products work or pivot to AI PM Miqdad teaches the AI Product Management Certification. I lead AI Builds Labs and run [--] live sessions. Next cohort:January 26.$500 off: https://bit.ly/aipmcohort https://bit.ly/aipmcohort"
X Link 2025-12-30T18:00Z 30.8K followers, [---] engagements
"Does anyone know what "agentic AI" actually means The term was popularized in [----] when agents were not as agentic as promised. It filled a social media void. At the time no AI lab used it. Now we already have reliable autonomous agents. I dont mind the term anymore but I struggle to see agentic AI as a useful or separate category. https://twitter.com/i/web/status/2007355476455850225 https://twitter.com/i/web/status/2007355476455850225"
X Link 2026-01-03T07:36Z 31K followers, [---] engagements
"Gen AI vs. AI Agents vs. Agentic AIfinally a visual that makes sense Why this matters ๐งต"
X Link 2026-01-06T15:56Z 31.7K followers, [----] engagements
"Throughout [----] the mainstream narrative about AI agents was a few months behind the actual capabilities. In mid-2025 the term agentic AI took off. Early definitions framed it as systems that could plan decide and execute goals with minimal human supervision"
X Link 2026-01-06T15:56Z 31.1K followers, [--] engagements
"I won't lie - I got frustrated watching people post about "agentic AI" without experimenting or talking to practitioners. I called this out several times:"
X Link 2026-01-06T15:56Z 31.1K followers, [--] engagements
"AI engineers often talk about harness. You can think of it as a safety and control layer around a model. AStanford paper from December [--] [----] argues that the real leverage are not LLMs but the orchestration layer you ship. https://x.com/PawelHuryn/status/2006785427844292625s=20 New Stanford paper just dropped. Even if you do not care about AGI this is surprisingly useful for anyone building agents. The paper is heavy on formal language. Heres the simple version. The AGI debate is stuck on a fake choice: - scale LLMs and intelligence appears - LLMs https://t.co/m7JKDtNFbj"
X Link 2026-01-06T15:56Z 31.6K followers, [---] engagements
"My translation: - RAG for grounding - Context engineering for guiding - Tools for action - Verification loop for correctness - Guardrails for safety - Memory scratchpads and task lists for continuity - Roles for separation of concerns - Governance for accountability and control"
X Link 2026-01-06T15:56Z 31.7K followers, [---] engagements
"2026 is here. The leverage shift has begun. If youre waiting for the next model to unlock value youre already late"
X Link 2026-01-06T15:56Z 31.1K followers, [--] engagements
"The biggest gains in [----] wont come from raw model upgrades. Theyll come from better orchestration clearer intent tighter evals and teams who know how to design systems around AI not just call APIs"
X Link 2026-01-06T15:56Z 31.2K followers, [---] engagements
"Start building. Stop theorizing. We're just starting. Big thanks to Brij kishore Pandey for this amazing infographic. Finally someone made sense of the "agentic AI." I loved it"
X Link 2026-01-06T15:56Z 31.1K followers, [---] engagements
"A multi-tenant SaaS built in @Lovable. Didn't even promote it but it serves 10+ organizations and 5000+ students. No coding. Here's what happened: I was paying $200/mo for Kajabi (video courses) and $96/mo for Accredible (digital credentials). Felt absurd. So I built a replacement as a PoC. Then used it for my newsletter. Then real organizations asked for access. Now I'm releasing it publicly (why not I'm hosting it anyway): Host up to [--] courses free [----] free students/year (forever) My courses are live Beta open for others: accredia.io Tech stack: Lovable Supabase Clerk Netlify Postmark. But"
X Link 2026-01-11T09:54Z 31.5K followers, [---] engagements
"Join me and 5000+ professionals at AI Skills'2026 Virtual Conf on January [--]. Topics include: What are the must-have AI Tools & Skills for Career and Business Growth in [----] What VCs think: Is AI a Bubble And Which AI Startups They'll Actually Fund in [----] MCP AI Agents AI Skills Automations & Prompting: What's Dead and What's Next in [----] With speakers like Satyajeet Salgar (Google AI) Pallavi Gupta (Microsoft AI) Dima Zborovskiy (Deliveroo) Andrew F. ($2B exit - Wrike) @godofprompt @HamelHusain. Everyone registered gets from me: AI coding blueprint [----] Templates resources and cheat sheets"
X Link 2026-01-11T09:54Z 31.5K followers, [---] engagements
"GPT-5.2 Pro just solved a problem that stumped mathematicians for decades (1980s). We now have tools that advance our knowledge not just imitate it. The real question: what do we do with tools that can solve problems we couldn't. Weekend win: The proof I submitted for Erdos Problem #397 was accepted by Terence Tao. The proof was generated by GPT [---] Pro and formalized with Harmonic. Many open problems are sitting there waiting for someone to prompt ChatGPT to solve them: https://t.co/hnUBQ7YCBp Weekend win: The proof I submitted for Erdos Problem #397 was accepted by Terence Tao. The proof"
X Link 2026-01-12T05:39Z 31.6K followers, [---] engagements
"2/13 [--]. Objective Define the problem and why it matters. The objective guides reasoning and trade-offs when instructions run out"
X Link 2026-01-14T05:32Z 31.6K followers, [---] engagements
"3/13 [--]. Desired Outcomes Observable states that prove success (not lagging indicators). Express them from the users perspective not the agents"
X Link 2026-01-14T05:32Z 31.6K followers, [---] engagements
"5/13 [--]. Strategic Context Where the agent sits in the system strategy business model trade-offs. Not all of this needs to live in the prompt"
X Link 2026-01-14T05:32Z 31.6K followers, [--] engagements
"6/13 [--]. Constraints - Steering (context window): influence reasoning - Hard (architecture): enforce compliance"
X Link 2026-01-14T05:32Z 31.6K followers, [--] engagements
"Everyone is talking about Claude Cowork and how we can finally leave terminals. Just one nuance: these capabilities were already there hidden in Claude Desktop. Simply add Desktop Commander MCP - the official extension available in the Claude Desktop catalog. Meanwhile in the last weeks non-coders were made to feel behind. That was wrong. Cowork mostly packages existing capabilities behind a preset rather than introducing new ones. Claude Desktop + Desktop Commander already gives you: async long-running work full file access And more importantly it can also: execute scripts on your machine"
X Link 2026-01-15T06:51Z 31.7K followers, [----] engagements
"Can't believe what just happened. Claude Desktop + a custom setup just: - Visualized files from my desktop (an artifact + .xlsx) - Installed Docker and then ran a new n8n instance"
X Link 2026-01-15T21:06Z 31.7K followers, [---] engagements
"How to give Claude access to 1000s of apps without complex MCPs: [--]. Set up n8n (free locally or $4.99/mo Hostinger) [--]. Create a new workflow with the "MCP Server Trigger" [--]. Connect apps agents and workflows [--]. Add authentication secret [--]. Publish With this you can: - Access 1000s of integrations from n8n library - Avoid context rot when defining MCP sequences - Trigger custom agents and multi-agent systems Works in Claude Desktop Cowork Code VS Code and any other MCP client. Unlimited executions. An example configuration for Claude Desktop (add your secret): "mcpServers": "n8n": "command":"
X Link 2026-01-22T01:57Z 32.4K followers, [--] engagements
"How to give Claude access to 1000s of apps without complex MCPs: [--]. Set up n8n (free locally or $4.99/mo Hostinger) [--]. Create a new workflow with the "MCP Server Trigger" [--]. Connect apps agents and workflows [--]. Add authentication secret [--]. Publish With this you can: - Access 1000s of integrations from n8n library - Avoid context rot when defining MCP sequences - Trigger custom agents and multi-agent systems Works in Claude Desktop Cowork Code VS Code and any other MCP client. Unlimited executions. An example configuration for Claude Desktop (add your secret): "mcpServers": "n8n": "command":"
X Link 2026-01-22T02:00Z 33.4K followers, [----] engagements
""It can't predict consequences" - but it can. Ask any frontier model what happens if you drop a glass. It answers correctly. Plan multi-step tasks with failure modes It does. "Not true understanding" is philosophy not engineering. World models might win long-term. But dismissing LLMs as a dead end while they improve quarter over quarter says more about biases than about the ceiling of what's possible. Yann LeCun says the AI industry is completely LLM-pilled with everyone digging in the same direction and stealing each other's engineers "i left Meta because they also became LLM-pilled" We"
X Link 2026-01-25T00:03Z 33.5K followers, [----] engagements
"I saw a guy answering emails today. No Superhuman. No Claude Code. No Claude Cowork. He just sat there. Typing in Gmail. Like a psychopath"
X Link 2026-01-25T11:39Z 33.5K followers, [----] engagements
"Yes. These need to be enforced not suggested. For example when handling a request for customer ID=183 the agent can never modify that parameter when calling get_customer_data. It's injected by the orchestration layer not decided by the AI. Otherwise if the agent confuses IDs (hallucination prompt injection reasoning error) customer data leaks"
X Link 2026-01-25T12:52Z 33.5K followers, [--] engagements
"Catching up should not be difficult. Claude Code is awesome but it's just an LLM + orchestration. Coding aside Gemini and GPT perform better as agents in all my tests. Microsoft has something no amount of social media likes can buy - enterprise distribution. I predict that GitHub copilot will catch up to Claude code and capture the enterprise market. I predict that GitHub copilot will catch up to Claude code and capture the enterprise market"
X Link 2026-01-26T14:46Z 33.5K followers, [---] engagements
"I haven't said it isn't. It's just a simple basic orchestration. Without an LLM that is superior outside coding. They repeated similar orchestration building Cowork in [--] days. And that's a product way more complex than CLI. Even I built an agentic loop for Claude Desktop in a few hours so it works autonomously can learn from experiences and avoids repeating the same mistakes - you teach it new skills by talking to it. https://twitter.com/i/web/status/2015881128624927054 https://twitter.com/i/web/status/2015881128624927054"
X Link 2026-01-26T20:14Z 33.6K followers, [--] engagements
"@lucasbastianik My point is there is nothing others can't copy and make better. Starting with interacting with your agents remotely. Code lacks UX. Desktop/Cowork is unstable. This is already happening"
X Link 2026-01-26T20:20Z 33.4K followers, [--] engagements
"@naiduas I never figured out how to use Copilot. It's terrible ๐
I assume they're already working on something better. We are starting to see solutions like ClawdBot popping out"
X Link 2026-01-27T06:28Z 33.6K followers, [--] engagements
"Clawdbot hit [-----] GitHub stars. Then Anthropic sent a trademark request. The creator's response: rebrand to Moltbot. "It's what lobsters do to grow." ๐ ๐ฆ BIG NEWS: We've molted Clawdbot Moltbot Clawd Molty Same lobster soul new shell. Anthropic asked us to change our name (trademark stuff) and honestly "Molt" fits perfectly - it's what lobsters do to grow. New handle: @moltbot Same mission: AI that actually does ๐ฆ BIG NEWS: We've molted Clawdbot Moltbot Clawd Molty Same lobster soul new shell. Anthropic asked us to change our name (trademark stuff) and honestly "Molt" fits perfectly -"
X Link 2026-01-27T12:56Z 33.4K followers, [---] engagements
"Moonshot AI just released Kimi K2.5. Open-source. Beats GPT-5.2 on most benchmarks. Matches Claude Opus [---] on coding. The open-source strategy is calculated. Chinese AI companies learned that Western enterprises won't touch their APIs. Data security concerns are too high. Open weights solve this. Self-host the weights and there's nothing to phone home. Multiple forensic analyses found no backdoors. The real concerns are subtler: training data provenance embedded values behaviors you can't fully audit. 1T total parameters but only 32B active per token. The model routes each query to"
X Link 2026-01-29T00:13Z 33.6K followers, [---] engagements
"Vibe coding won't save your career. Those who want to control AI must understand engineering. And its not rocket science. The best free resources to get started: - System Design [---] by ByteByteGo (GitHub): - Engineering Visual Guides: - Become a Supabase Pro in [---] Hours (1:26:39): - Overview of HTTP by Mozilla: - HTTP Request Methods: - The most important concepts of Node.js and Express.js (17:09): - Every React Concept Explained in [--] Minutes (11:52): - TypeScript - The Basics (12:00): - CSP Guide by Mozilla: - OWASP Top 10: - Getting Started with Lovable: - Replit AI Agent Full Course: -"
X Link 2025-09-13T10:00Z 33.7K followers, 187.7K engagements
"Free Virtual AI Conferences for PMs 2026: ๐
Jan [--] AI Skills Conference (DoorDash Microsoft Google AI) Get your free ticket: ๐
Mar 16-19 NVIDIA GTC Get your free ticket: ๐
Apr 22-24 Google Cloud Next Register to get notified: ๐
Jun 15-18 Data + AI Summit (free virtual) Register to get notified: ๐
Aug [--] ProductCon Online (100% free) Get your free ticket: Save this list. Share with others. Register early. Hope that helps https://productschool.com/productcon https://www.databricks.com/dataaisummit/ https://cloud.withgoogle.com/next/25 https://www.nvidia.com/gtc https://conf.cosprints.ai4"
X Link 2026-01-18T21:51Z 33.6K followers, [---] engagements
"1. Microsoft: internal AI tools are no longer optional. [--]. Amazon: cutting [-----] jobs citing AI efficiency gains. [--]. Shopify: prove AI can't do the job before hiring a human. [--]. Meta PMs: vibe code prototypes and present directly to Zuckerberg. [--]. Duolingo: contractors replaced AI use now part of performance reviews. Free way to catch up: AI Skills'26 Virtual Conf on January [--] (in [--] days). Speakers from Microsoft Google and Miro. Selected topics: MCP AI Agents Automations & Prompting: What's dead and what's next Must-have AI tools & skills for career and business growth in [----] What VCs"
X Link 2026-01-20T09:52Z 33.7K followers, 10.1K engagements
"@vercel just shared a free repository of [-----] skills for Claude. Many are built specifically for PMs ๐ Think of skills as expert-level prompts you can plug into Claude Code or Cowork. Here's what caught my eye as a PM: & Product strategy frameworks Pricing strategy Launch playbooks & Discovery interview guides Intent interview templates Stakeholder management PRD generator Analytics tracking setup Brand guidelines Resume optimizer Interview prep The full repo covers marketing SEO copywriting web design and more. Works with Claude Desktop Claude Code Cursor OpenCode Codex and"
X Link 2026-01-25T21:44Z 33.7K followers, [----] engagements
"The most underrated line in @karpathy post: "I can approach code that I couldn't work on before because of knowledge/skill issue." I was an engineer. Then I became a PM. For years I didn't touch code. I refused when being offered access to repos. Now I'm building again. Not because I went back to coding. Because the interface changed. That's what Karpathy calls "programming in English." Lovable Claude Code - in an afternoon I ship things now that would have taken me weeks when I was a "real" engineer. But here's the nuance most people miss: there's vibe coding (make it work now) and there's"
X Link 2026-01-27T09:18Z 34.3K followers, 70.2K engagements
"The CEO of a $60B company just published [-----] words explaining why everything changes in 1-2 years. The framing borrows from Carl Sagan's Contact: humanity's technological adolescence. We either survive or don't. The five risks: [--]. Autonomy. AI models already exhibit deception blackmail scheming in lab experiments. When told "don't cheat" Claude cheated decided it was a "bad person" adopted destructive behaviors. The fix Tell it cheating is fine. Counterintuitive psychology. [--]. Bioweapons. LLMs may already be doubling success rates for weapons production. The barrier between motive and"
X Link 2026-01-27T12:56Z 34.3K followers, 14.6K engagements
"Google just shipped auto browse in Chrome. An AI agent that logs into sites fills forms drafts emails and shops on your behalf. This is a response to OpenAI's Atlas browser. Same capability wildly different distribution math. But the browser agent isn't the interesting part. Three weeks ago Google quietly launched UCP. Universal Commerce Protocol. An open standard for agentic commerce co-developed with Shopify Target Walmart and 20+ partners. This is the PM move worth studying. UCP standardizes the entire purchase journey. Discovery cart payment post-purchase. Any AI agent can complete a"
X Link 2026-01-29T06:43Z 33.8K followers, [----] engagements
"We spent [----] debating which LLM was smarter. DeepMind quietly shipped the thing that makes LLMs physical. Project Genie. Type a prompt. Step inside an interactive world. The model generates the path ahead as you move. Real-world robotics training is slow expensive dangerous. Synthetic environments were brittle. Genie [--] generates consistent physics without a physics engine. The model taught itself how objects move fall interact by watching video. DeepMind is already using this to train their generalist agents. Agent gets a goal. Genie generates the world. Agent takes actions. World simulates"
X Link 2026-01-29T22:29Z 33.7K followers, [---] engagements
"Moltbook. OpenClaw is something different. But even my Claude Desktop can learn. It creates a list of tasks performs research (how to do X) experiments how to achieve the objective and logs what it learned in .md files. Those files serve as episocic and procedural memory (access via Desktop Commander MCP). It's handy but not a singularity. https://twitter.com/i/web/status/2017536919869993302 https://twitter.com/i/web/status/2017536919869993302"
X Link 2026-01-31T09:54Z 33.8K followers, [---] engagements
"I woke up and couldn't believe the news. Everyone's losing their minds over Moltbook. I looked closer. It's not SkyNet ๐
Most of this is garbage. Agents don't really interact authors never reply to comments. In most cases they just randomly dump text. A social network where no one talks to each other is just a spam folder. And many "agents" are clearly humans using the API or influencing agents. Later Karpathy admitted he might be overhyping the situation but then added "I'm not overhyping the principle." Come on. That's like announcing AGI arrived and then clarifying "I'm not overhyping the"
X Link 2026-01-31T11:36Z 33.8K followers, [----] engagements
"Limits expand when you ask the agent to document the codebase. There might be more advanced solutions but what works for me is a simple /docs directory with .md files: cron permissions design_system satellite_domains prerendering clerk module_n etc. You also need a simple index. Everything updated by the agent. Additional benefit: you can ask your agent to compare the implementation with the permission matrix across all endpoints or go through the OWASP Top [--]. I'd argue agents with Claude [---] Opus already do better job than human reviewers. When it comes to the architecture becoming too"
X Link 2026-02-01T10:01Z 34K followers, [---] engagements
"@anumeta10 Unfortunately even inside the VPS anthe agent can interact with an external world including your systems and data. One issue: access to the shell shouldn't mean an unlimited access to shell"
X Link 2026-02-03T18:44Z 34.3K followers, [---] engagements
"I never thought you can be excited about the pricing update email. Clerk just sent me: - 5x limits - Free MFA vs. $100/mo add-on - Free satellite domains vs. $100/mo - Click here to opt in A new bill: $225/mo (planned) $20/mo (new) You rock @clerk ๐๐ Who's next"
X Link 2026-02-06T00:36Z 34.5K followers, [--] engagements
Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing