Sottiaux's DM revelation (4217 likes): 'I asked Codex to pull up some stats and I receive on average one DM or email every 6 or so minutes to ask for a reset. I occasionally do oblige if it comes with really solid feedback or banter.' Plus his /goal recommendation (2637 likes): 'I recommend you explore /goal in Codex, it's a pretty powerful loop with GPT-5.6 Sol.'

Steipete's table rendering (1382 likes): showing beautifully rendered tables in what appears to be a presentation context. Plus his e2e testing setup (399 likes): 'I gave codex a video-enabled remote KVM so it can automate e2e test the iMessage-integration on OpenClaw. iMessage is unreliable in VMs, and certain features such as read receipts require SIP to be disabled.'

Garry Tan on AI detection futility (418 likes): 'Can't wait for the AI to get so good none of this business about detecting AI matters anymore. Silverware used to be handmade, but nobody complains about their dinner fork is stamped by a machine. The quality of ideas matter.' Rauchg on tweets (286 likes): 'Writing a banger tweet is AGI-complete.' Plus infinite compute (222 likes): '10,000 concurrent + 5,000 CPU cores per minute and these quotas are raisable.'

Levie on enterprise tokens (133 likes): '99% of tokens in the world will get consumed in an enterprise context. Code, life sciences research, manufacturing, security, fraud detection.' Dan Shipper on Google (155 likes): 'Google needs to catch up on frontier coding. Demis believes different fundamental research directions like world models are more important to his long term goal even if less important competitively today.' Nan Yu's dumb question (118 likes): 'How is ChatGPT not an agent?'

Peter Yang's /human-review skill (179 likes): a tool to edit HTML and Markdown files directly with a visual editor inside Codex/Claude Code. Plus his Luna Extra High usage note (70 likes) and PM-turned-PR-reviewer joke (32 likes). Nikunj's AI jargon forecast (28 likes): 'out of distribution, control plane, unverifiable fields, rails, intelligence per watt, cope, angst.' Plus his Nikita/Jeff Dean departure note (33 likes) and 2026 AI startups meme (16 likes).

Swyx on multi-agent kanban (7 likes): a primitive form of multi-agent AGI where threads ping back when done, creating implicit dependency graphs. Madhu Guru on slow AI diffusion (57 likes): 'We greet users with a blank window and ask them to write a prompt. Then pick a model, decide whether they need an agent, know what context windows are.' Plus his Jeff Dean tribute (20 likes). Google Labs Dreambeans expansion (406 likes): personalized daily stories for AI Ultra and Pro users.

MAD Podcast: Matt Turck (FirstMark) interviews Mitch Troyanovsky, co-founder of Basis (unicorn AI accounting company). Basis agents run autonomously for hours/days, completing tasks like entire tax returns end-to-end. Key topics: why accounting is a compression/intelligence problem over the economy, the Memento analogy for LLM memory (large working memory, no short/long-term memory by default), why BabyAGI failed (models weren't smart enough at long context), the three holy shit moments (Opus 3, o1, o3), process supervision vs outcome rewards, evals vs real-world production (100 evals passing doesn't mean production-ready), the English context is more precious than code context, and how reasoning models let agents self-heal by titrating compute per step.