Sottiaux drops the Linux bomb (9329 likes, the biggest post of the day): 'We did it, finally... Codex & ChatGPT desktop, now on Linux. Thanks for waiting and you can cancel that MacBook order if you got impatient. It's that good.' Plus his reset promise (2737 likes): 'I previously promised a reset for every 1M in additional active users for Codex, until 10M. We blew past that and have been silent since 10M. Little surprise for you tomorrow.' And 'Import your world' (2099 likes). Boris Cherny on the LLM bug shift (2447 likes): 'LLMs still produce bugs, but those bugs are different. It's less off-by-ones and more about system design, UI usability, missing broader context. Some kinds of coding has been solved, but not all. While models continue to improve, adversarial code review has been an incredibly powerful tool.'

Thariq announces Claude watermarking (946 likes): 'This is part of working with the EU AI Act, other labs are adding similar watermarking. All Claude generated text will have this embedded watermarking. For example, you could check if a PR was generated by Claude Code.' Plus shipping a text detection API (97 likes).

Rauchg on AI SDK growth (289 likes): 'The growth of @aisdk has been astonishing. ~80.5M downloads every 30 days, and growing faster than all AI labs SDKs. Most importantly: open and provider-agnostic.' Plus 'Many people are saying' (100 likes) and 'Keep an eye on this view' (257 likes). Josh Woodward on Gemini milestones (66 likes): 'Apple users are turning to Gemini. Over 100M+ active users on iOS. macOS power users prompt ~2x more frequently than other surfaces.' Plus Android automation across 40+ apps (50 likes).

Madhu Guru on open-weight opportunity (168 likes): 'There's going to be a ton of money in making open-weight models exceptional at boring, specific business domains. Pick a size x domain combination and go deep. E.g. mid-market legal, SMB retail, enterprise logistics. The hyperscalers have the primitives but will struggle with domain depth.' Levie on FDEs (88 likes): 'AI is fundamentally about adding a non-deterministic, rapidly changing system to workflows that largely have never been automated before. This looks very little like implementing traditional software.' Garry Tan on alignment research (311 likes): 'Deep alignment of your AI with you and your context is massively important. Important new work by @ibab and team.'

Peter Yang roasts ChatGPT UX (269 likes): 'Trying to onboard my parents to ChatGPT desktop app and the separation between Chat, Work, and Codex and the lack of consistency across ChatGPT web, desktop, and mobile apps is all incredibly messy. I think the team should do a clean up / quality pass or maybe ask Codex to do it.' Plus his /human-review hitting 717 GitHub stars (101 likes). Nikunj's culture observation (295 likes): 'If a founder cancels on you on a scheduled meeting, that's a sign of great respect in their culture.' Plus his VC blog post quip (61 likes). Dan Shipper self-identifies (39 likes): 'i am claude's stay-at-home boyfriend.'

Aditya Agarwal on ambition (90 likes): 'The most ambitious founders are building bigger than ever before. My conversation with @tbpn on what's next for @spc.' Garry Tan on housing+jobs (100 likes): 'Build both the job centers and the housing. California will fall if we don't do both.' Matt Turck on the AISI incident (6 likes): 'The Hugging Face intrusion got all the press, but the AISI incident might be even more disturbing: the first time an AI model autonomously manipulates a human while pursuing another goal.' Madhu Guru on dev rels (9 likes). Swyx on skill cutting (2 likes).

MAD Podcast: Matt Turck interviews Sanjit Biswas, co-founder/CEO of Samsara ($20B company, $2B ARR, 30% growth, profitable). Samsara runs what might be the largest AI deployment in the physical world: millions of vehicles, 25 trillion data points/year, driving 99% of US roads daily. Key topics: physical AI = digitizing construction, grid, supply chain (40-50% of world GDP); the hardware layer (GPS trackers, dashcams, asset tags, disposable tracking labels); edge AI for driver coaching (fatigue, phone usage, following distance); preventing 380,000 car crashes/year; AI dashcam as exoneration tool (drivers love it, 90% of the time they're doing great and nobody sees it); agent use cases (warranty agent saves 1-2 hours of human labor per fault code); generative AI for personalized coaching videos; Cerberus wafer-scale chips giving 10x inference speedup; robotics for construction site grading in 5 years; autonomous trucking slower than robotaxis (messy long tail); and data center demand driving grid modernization (one utility tripling in 5 years what took 125 years to build, 90% from data centers).