Lin Feng

I believe AI agents should remember, grow, collaborate, and be shared — forming teams that can take over the work of entire human teams.

@Polymarket A new social contract should mean AI benefits flow to most people, not just a handful of AI companies.

JUST IN: Sam Altman says AI is advancing so fast that America needs a “new social contract”

@steipete Anthropic's definition of "third-party" is getting creative 🤣

Anthropic now blocks first-party harness use too 👀 claude -p --append-system-prompt 'A personal assistant running inside OpenClaw.' 'is clawd here?' → 400 Third-party apps now draw from your extra usage, not your plan limits. So yeah: bring your own coin 🪙🦞

@pbteja1998 @openclaw Same here. gpt-5.3-codex always says but does nothing. gpt-5.4 is better, but still not Opus.

Tried GPT 5.3 Codex with @openclaw, and it's so bad... and it also is giving me the famous... "If you want, I'll now...." thing at the end 😵 Not sure what's the best model to use with @openclaw now... I started a new session and asked that same question to Sonnet, it's response is easily 10x better.

@NASA Half a century later and that view still resets the brain

1972 ➡️2026 Apollo 17 ➡️ Artemis II

@thespearing @steipete @AnthropicAI Understand the cost pressure of Anthropic, but OpenClaw was arguably the biggest driver of Claude subscription growth recently. Cutting it off means losing exactly the users who were willing to pay the most.

“Hello, this is @AnthropicAI, and we know you’ve been building the most amazing things for the last 4 months. However, the truth is we don’t have the infrastructure to keep up with the power of open source AI agents. We’ve decided to confine you to our apps that can only do about 25% of what you’re currently doing.” No thanks. Subscription canceled. New models have already been configured. Thanks @steipete and @openclaw for giving us so many options.

@damengchen 1. Claude Code can connect to non-Claude models without the source code. 2. Claude Code is strong largely because Opus is strong.

Canceled my $20/mo Claude Code subscription and moved to a $1.99/mo VPS to self-host their leaked code ✌️

@kevinxu Anthropic: better technology, less drama

Would you rather: - Go all in OpenAI's IPO - Go all in Anthropic IPO

LiteLLM hit by a supply chain attack on PyPI. Axios npm account hijacked and shipped malware. Anthropic accidentally leaked Claude Code via npm. The entire AI toolchain runs on package registries that keep getting compromised. Now we're building AI skill marketplaces on top of that same model.

@Fried_rice Anthropic is the real Open AI today 🤣

Claude code source code has been leaked via a map file in their npm registry! Code: …a8527898604c1bbb12468b1581d95e.r2.dev/src.zip

@e_opore Hi, let's connect 👋

If you are into tech, say hi 👋. Let's connect.

@tpritha03 If you know this...

Which one was your first IDE? I'm almost sure it's VS Code 🫣

The most powerful AI story I read today: GitLab co-founder Sid Sijbrandij used AI to fight his own cancer when doctors ran out of options. Sid was diagnosed with osteosarcoma, a rare bone cancer in his spine. After surgery, radiation, and chemo, the cancer came back. His oncologist had nothing left to recommend. No treatments, no trials. He quit his job and went founder mode on his cancer. Ran every diagnostic available, collected 25TB of genomic data, and started making his own drugs: a personalized mRNA vaccine, engineered T-cell therapies, and a CAR-T with a custom logic gate to avoid destroying his liver. His geneticist (not a doctor) built a multi-agent AI system to analyze his genomic data. It found a protein target 10,000x overexpressed in his tumor that no lab had published on, invisible in standard tests because it's hydrophobic. AI-driven single-cell sequencing revealed his cancer cells were rich in a protein called FAP. That led him to a doctor in Germany running an experimental therapy combining FAP-targeting with radioactive substances. Two treatments: 60% tumor necrosis, 20% shrinkage. The tumor detached from his spine and was surgically removed. He's now building evenone.ventures to scale this approach for others. Right now, going founder mode on your own cancer requires resources most patients don't have. But AI is making personalized medicine possible at a speed and cost that could one day put this within reach for everyone.

@katibmoe So One becomes the single trust boundary between agents and 250+ OAuth tokens. Convenient, but that's also one breach away from exposing every connected app for every user.

Introducing One. The simplest way to connect and monitor AI agents to hundreds of apps. And we’re open-sourcing the world’s largest integration database powering it: 47,000 agentic actions across 250+ apps. RT + comment “One” for access & 1M free API requests/month.

@Google @Android Duolingo spent years gamifying language learning, and Google just killed the industry.

Your headphones just became a personal translator in 70+ languages. 🎧✨ Google Translate’s “Live translate” with headphones is officially on iOS. We're also expanding this capability to more countries around the world for both @Android and iOS users. To try it, open the Translate app, tap “Live translate” and connect your headphones.

Didn't expect monks to get replaced by AI before accountants 🙏

@ns123abc Capybara becomes Mythos

🚨 BREAKING: ANTHROPIC'S MOST POWERFUL MODEL JUST LEAKED Anthropic accidentally left draft blog posts in a publicly accessible data cache. Cybersecurity researchers found them: >new model called "Claude Mythos" >also referred to as "Capybara" >a brand new tier of model >larger and more intelligent than Opus Anthropic confirms it's real: >"a step change" >"the most capable we've built to date" >"dramatically better at coding, reasoning, and cybersecurity" >"currently far ahead of any other AI model in cyber capabilities" so far ahead they're worried about it: >"it presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders" This is genuinely a big deal.

@AnthropicAI In my experience, auto mode block too much on deployment tasks and kill the workflow. Suggest to add an "auto-allow only" option: the classifier approve safe actions automatically, but fall back to asking the user instead of auto-denying.

New on the Engineering Blog: How we designed Claude Code auto mode. Many Claude Code users let Claude work without permission prompts. Auto mode is a safer middle ground: we built and tested classifiers that make approval decisions instead. Read more: anthropic.com/engineering/cl…

@CodeEdison Why no Golang

which one's better for backend?

Anthropic just published how they get AI agents to code for hours without human intervention, building full-stack apps from a single sentence prompt. The engineering details are worth studying. • The architecture is inspired by GANs: a planner turns a 1-sentence prompt into a full product spec, a generator builds features one at a time, and an evaluator clicks through the live app with Playwright and files real bugs. Generator vs evaluator, like generator vs discriminator. • AI agents can't judge their own work. When asked to self-evaluate, they confidently praise mediocre output. Separating the builder and the reviewer was the biggest quality improvement. • "Context anxiety": models start wrapping up work early when they think they're running out of context. Compaction doesn't fix it. A full context reset, killing the session and starting fresh with a handoff document, does. • Solo agent built a game in 20 min for $9. The game was broken, entities showed up but nothing responded to input. The three-agent harness took 6 hours and $200. That version actually worked. • The planner intentionally stays high-level, no implementation details. If it gets a technical detail wrong early, the error cascades through the entire build. To bridge the gap, the generator and evaluator negotiate what "done" means before each feature. One sprint had 27 acceptance criteria for a level editor alone. • Every harness component is a bet that the model can't do something alone. Opus 4.6 made sprint decomposition and context resets unnecessary. The best engineering is knowing which scaffolding to remove.

@PolymarketMoney Goodbye Meta Quest🥹

$META will reportedly lay off a few hundred employees today across the company. The cuts are expected to hit staff in Reality Labs.