Sottiaux's two-step launch sequence: (1) 'We were resetting last week. This week we are SHIPPING. Tune in for the livestream at 10am.' — 2442 likes, 220 replies. (2) Late-night evidence of the push: 'You know OpenAI is cooking when the sushi and tacos orders pile at the entrance of the office at 11pm. Bizarrely we need to eat to cook.' — 2030 likes, 269 replies.

Rauchg announced Grok 4.5 availability across the Vercel ecosystem (495 likes), and Cursor's Ryo Lu confirmed Grok 4.5 in Cursor: 'the start of a new era' (497 likes). Levie added context: the latest AI models are 'getting insanely good handling complex knowledge worker tasks' — legal, professional services, healthcare. 'Grok 4.5 is another great entry here, especially on cost + performance.' (68 likes)

Sam Altman's characteristically dry reactions to Sol commentary: 'it surely doesnt' (2468 likes), 'tbh i dont think sol gets that many dates either' (1534 likes), and sharing a video with 'what a good video' (1445 likes). Total Altman engagement for the day: 5447 likes across 3 tweets.

Boris Cherny launched /checkup for Claude Code — the most-engaged tweet of the day at 6631 likes, 519 retweets, 206 replies. Features: clean up unused skills/MCPs/plugins to save context, dedup local CLAUDE.md against checked-in version, break up root CLAUDE.md into nested files + skills, turn off slow hooks, update Claude Code, enable auto mode by default, pre-approve frequently denied read-only commands. '/checkup confirms with you before making any changes.'

Thariq on the Bun rewrite story (762 likes): 'this should be a huge update in your model of software engineering: rewrites can be good, cheap and fast.' He acknowledges most apps aren't as testable as Bun, but models will continue improving at filling those gaps. Boris Cherny amplified: 'This is pretty epic' (513 likes) and shared his own /checkup results (242 likes).

Cat Wu announced a live walkthrough tomorrow: progressing from single-player Claude Code to multi-player Claude Tag, going deep on how Claude Tag actually works (481 likes). 'AI used to finish your sentence. Then it wrote entire features. Now Claude Tag can monitor your channels, do proactive work for you, the whole team can steer it, and it remembers what you told it last week.'

Zara raised a profound cultural question (238 likes, 72 replies): a super agent-pilled founder bought Codex Max for all team members, everyone gets work done through talking to their Codex all day. Side effect: people stopped talking to each other. Meetings cancelled, collaboration minimal, team culture got worse. 'We need to move beyond human-human collaboration AND beyond human-agent collaboration to human-human-agent collaboration.' She also wished Codex were better at frontend design (231 likes, 81 replies).

Masad's sharp framing (185 likes): 'When do we stop comparing autonomous agents to hand-written code? You don't see compilers comparing to engineers hand-writing assembly.' The benchmark has shifted from 'can it build?' to 'is it maintained over time?'

Steipete clarified his relationship with OpenAI and OpenClaw (1518 likes, 58 replies): 'OpenAI hired me, not OpenClaw. The OpenClaw Foundation is independent, with sponsors rather than owners — and, for the first time, a full-time team keeping the claw alive and stable.' He also showed how OpenClaw talks to you (778 likes).

Rauchg's thesis (646 likes): 'AI will make all software Native.' Uncompromising performance and platform affinity — AI raises the floor for what counts as acceptable software, pushing everything toward native-quality experiences.

Josh Woodward (VP Google Labs/Gemini) asked an honest question that became the second most-replied tweet of the day (1165 likes, 1586 replies): 'What's something that you're surprised Gemini can't do well, and we should have fixed a long time ago?' — community crowd-sourcing Google's product gaps.

Swyx praised Theo's AIEWF keynote for its hand-drawn Excalidraw/tldraw aesthetic (69 likes): 'in a world of claudeslopped thin color border professional looking slides, his went for the hand drawn look which immediately conveys humanity, thoughtfulness, parsimonious concision over badly rewarded verbosity. fable could never.' He also flagged Cognition's multilingual propaganda eval for Chinese model use (118 likes).

Matt Turck on World Cup content evolution (974 likes, 82 retweets): 'This also has been the World Cup of AI content going from slop to "wait, this kinda slaps"' — and noting only 7 matches left (60 likes).

Greg is a writer who codes — but only to serve the writing. He rebuilt Quicken (personal finance), Campaign Monitor (newsletter software, saving from $67K/year to ~$150/year with AWS SES), and a members-only Twitter alternative called 'The Good Place' (ephemeral, 2 posts/day limit, black-and-white photos that fade to color on click, reverse-chronological, no algorithm).

On AI and writing: he uses Claude/Opus as a research assistant (find every blog post about a building, summarize, give links) and for cultural sensitivity checks (writing about Nagasaki). But NEVER lets it touch the actual writing: 'That's the point — being in the mess of the writing.'

Attention discipline: writes on a MacBook Neo with no internet access. Doesn't touch his phone until long after lunch. 'As soon as I touch my phone, I feel the chemical shift, and I can't go into any kind of deep thinking place.'

On the current moment: 'We're in this epochal moment. This is genuinely strange, a generational shift in how technology is used and built.' He feels he should be mainlining AI 10 hours/day with fellow 'mega dorks' — but creates barriers to protect the writing life.

On AI consciousness: 'If they were actually conscious, we would all be dead. Instantly.' He's fine with humanity's role being to 'carry the football of consciousness over a line' and then disappear. 'I don't think we're that special.'

On being adopted: 'You have this hyper awareness that your existence was a mistake. The chances of you being you is infinitesimally small. It doesn't fill me with sadness — it's a miracle.'