Peter Yang flags a shocking policy: 'Suspending Fable for all foreign person inside the US is wild. How can you even enforce this properly?' 2015 likes.

This refers to Anthropic restricting access to Fable for non-US-persons within the United States, likely due to export control concerns around the Mythos-class model's capabilities.

Levie calls this a big turning point for AI regulation: 'The government is starting to deem some models too powerful for certain uses, which creates a precedent for a range of possible controls in the future.' He's in the camp that this is unnecessary and we should regulate use of AI, not underlying models. 'Either way, it's unlikely that we're going back to a world where the government doesn't have far more meaningful involvement in the rate of AI progress.' 451 likes.

Peter Yang's prediction in response to the Fable access restrictions: 'I think ID verification will soon be required to access the best models.' 674 likes.

This is a logical consequence: if models are subject to export controls, identity verification becomes the gating mechanism for access.

Thibault Sottiaux (OpenAI) responds to user feedback about Codex usage resets being annoying without warning: 'Next time we press the button you will get to choose when it actually applies. Happy codexing.' 5271 likes.

A small but meaningful product improvement—turning a surprise reset into an opt-in experience shows OpenAI is listening to the power-user community.

Amjad (Replit) reveals a principled business decision: when enterprise customers asked for a token consumption leaderboard during the tokenmaxxing craze, Replit said no. 'Would've been great for business but we're not in the business of selling tokens for the sake of tokens. We sell outcomes. And we knew it wouldn't last.' 228 likes.

This is a counter-position to the tokenmaxxing trend—betting that quality of outcomes, not volume of tokens, wins in the long run.

Steipete reports: 'I can barely keep up with implementing/testing/landing all the Issues/PRs folks submit. Codex runs INSIDE crabbox while it is building crabbox. Codex been looping nonstop for the last 4 days in multiple trees. Since all of it is e2e verifiable it basically builds itself. Codex even signs up for the services automatically via browser/computer use. My main job is adding credit card details and closing things that I don't see as a fit.' 234 likes.

The meta moment: Codex building the tool that Codex runs inside. The human's job has been reduced to credit card details and product judgment.

swyx's provocative thesis: after PRs die and Code Review dies, does Git need to die next? Roughly 20-40% of code spend is just managing merge conflicts. 'We don't do line by line merge conflicts when we collaborate with human colleagues—instead we chat, suggest edits, do side comments, and an owner ships it.' Maybe the future codebase looks more like a Notion or Linear database than .git objects. 'It will be less efficient, but more scalable. Exactly the Salty Lesson.' 90 likes.

This is a direct challenge to the fundamental unit of version control. If agents produce and review code autonomously, the merge-conflict abstraction may be obsolete.

Rauchg announces DarknessAgent: a unified abstraction to orchestrate and integrate any agent's brain into your app. The Vercel AI SDK now frees you from both model lock-in and agent lock-in. 828 likes.

This is Vercel positioning the AI SDK as the middleware layer between application developers and the proliferating agent frameworks—owning the orchestration layer rather than the agent itself.

Alex Albert (Anthropic) shares a practical Fable prompting tip: 'Fable feels superhuman at working over long agentic conversations, sometimes to the point where I can't keep up with what it's telling me. This prompt snippet has been the best fix I've found for getting it to write clearly and drop any jargon.' 880 likes.

The prompt snippet instructs Fable to communicate in plain language without technical jargon—a useful pattern for anyone whose model output has become harder to follow than the code itself.

Garry Tan's most-shared thought of the day: 'Everyone thinks AI coding tools set founders free. Watch what people actually build with them: rules, approvals, process, layers. The same cage, assembled faster. The tool that can scaffold anything in an afternoon will scaffold your bureaucracy in an afternoon too. Speed of construction is speed of calcification. Build the thing that lets you create new things: experiences that didn't happen before.' 746 likes.

This is a profound insight about the difference between building faster and building differently. AI tools don't just accelerate existing workflows—they also accelerate the creation of process debt if you're not intentional about what you're building.

Rauchg declares 'HTML is so back' with a drag-and-drop web creation demo. 1065 likes.

The continued resurgence of HTML as the lingua franca of AI-generated interfaces—every model can produce it, every device can render it.

Amjad announces a Replit incentive program: if you make money on Replit, you get free credits. 100 likes.

A clever alignment of incentives: the platform wins when creators monetize, so remove the compute cost barrier for those who are already generating revenue.

Unsupervised Learning's AI Vibe Check with Ari (Datalogy, ex-DeepMind/Meta) and Rob (Radical Ventures). Key themes: (1) Open weight models are dying—Meta pulling back from open source, Chinese labs keeping best models proprietary behind APIs. 'There's no business model for open source AI,' says Rob. Ari agrees: 'We probably hit the peak number of open models and it's now going to get less and less.' (2) The spicy prediction: labs may shut down their API business as compute crunch intensifies, prioritizing first-party products. (3) Token costs are becoming a real problem: Ari reports enterprises seeing bills double overnight from model upgrades, driving strong demand for cheaper alternatives. (4) Coding agents crossing thresholds: engineers shifting from ICs to managers of agents. The bottleneck has moved to code review and understanding, not production. (5) SaaSpocalypse narrative: Rob thinks it was rational for many companies but overdone as a blanket statement. 'There's no way one or two or three companies will win every single important market.' (6) Compute crunch implications: drives efficiency innovation, benefits non-NVIDIA chip makers, pushes toward smaller models and distillation. (7) On RSI: Ari is a believer in recursive self-improvement but doesn't buy the exponential takeoff narrative—compute is the fundamental limiter.