Altman's growth confirmation (9749 likes — today's most engaged tweet): '5.6 sol growth is insane. The inference team has done heroic work to support demand. We are going to move mountains to continue.' He also backed open-source harnesses (4800 likes) and posted a simple 'hello!' that got 5366 likes.

Sottiaux at the 9M threshold: 'Embarrassment of riches. But looks like we might hit 9M soon. Should we reset the ChatGPT Work and Codex usage again or give it some space?' (3608 likes). Then pivoted to a creative promotion: 'Or… what if we gave you $100 in Codex credits if you tell us what you love about GPT-5.6 Sol or why you switched?' (5757 likes). His third tweet: just scrolling Twitter looking for feedback (5530 likes).

Peter Yang announced a new video tutorial dropping tomorrow on how he uses ChatGPT Work (aka Codex) for 'almost everything on my computer' (36 likes). Aditya Agarwal noted the feature richness vs. friction tradeoff: 'I used ChatGPT (Legacy) 15-20 times a day for quick things... now it's more powerful but slower for simple tasks' (18 likes).

Anthropic's Katelyn Lesse & Angela Jiang (Training Data & Ecosystem team) on moving from 'walled garden' to 'ecosystem': 'Building an ecosystem, not a walled garden' (podcast title). Key themes: three-layer abstraction (knowledge → execution → coordination), training data strategy, investing in partner infrastructure rather than competing, and the tension between open and closed. They discuss how Anthropic uses synthetic data, partners for scaling, and where to draw lines on openness.

Anthropic launched Claude for Teachers — purpose-built for K-12 education with privacy-first design. Ask for a lesson plan and Claude starts from your state standards and high-quality curricula via Learning Commons integration (647 likes). Key privacy guarantee: never trains on conversations, student data protected by data residency (514 likes). The move signals Anthropic is expanding beyond developer tools into education.

Levie on why code is agent-friendly (149 likes): 'One of the many properties that code has that makes it highly amenable to agents is that you can more or less quickly test it.' Code is one of the few domains where you can automatically verify correctness — which is why coding agents work better than agents for most other domains. He also backed a standards body for AI (112 likes).

Ryo Lu's reflective thread (809 likes, 49 retweets): 'when the dream becomes the job — i used to think the dream was to make my hobby my work. to spend my life close to the things i loved: computers, code, the craft...' A meditation on the transformation that happens when passion scales into responsibility. He's also assembling 'the best design team' at Cursor, looking for design engineers too (896 likes).

Rauchg opened up the AI token flows dataset from Vercel AI Gateway (83 likes) — 'fascinating insights contained within!' He also promoted agentmail: tell your agent to vercel install agentmail — no signup, automatic setup, unified billing (159 likes).

Steipete shared two practical tips: 'That's why you always wanna run autoreview' (141 likes) — showing a code review catch. And: 'Suno AI is delivering bangers!' (167 likes). Plus a note that stress-testing burns tokens but calms nerves (13 likes).