Sottiaux's Sunday triple (7425 likes, top of the day): 'Good Sunday. Reset has been propagated to accounts and we landed some fixes to usage for things mentioned yesterday as issues we found. You should feel a positive difference. More to come tomorrow and will keep communicating.' Plus his thesis (5442 likes): '2026 is the year companies start seriously caring about model efficiency and reliability as it becomes critical infrastructure.' And the charming correction (2474 likes): 'Meant 2pm obviously.'

Levie's retirement meme (1985 likes): 'I'm good to retire now' — captioning a moment presumably too perfect to top.

Rauchg on extending fx with open protocols (703 likes): 'MCP (modelcontextprotocol.io), Skills (agentskills.io), Plugins (agent-plugins.org). And the best one, Unix: ① Small programs that do one thing well and compose by calling other programs. ② libfx that enables embeddability into more complex programs. You should be able to build your own CLI, background agent, software factory. Local or cloud.' Plus his elasticity observation (325 likes): 'Intelligence is getting cheaper. OpenAI Sol's price reductions & discounts on Vercel AI Gateway have made Sol our fastest-growing frontier model. The demand for intelligence is highly elastic: as inference costs fall, usage grows rapidly.'

Garry Tan's prediction (567 likes): 'systems of record will need to become AI harnesss or face replacement by agents.' Nan Yu's deadpan (5 likes): 'Also every other category of software.'

Ryo Lu's farewell poem (808 likes): 'leaving / leaving is the process from fullness into emptiness / from the familiar circle of life / the routines undone / the streets i knew by heart / the people i kept meeting / the names and labels that had gathered on my skin / toward a place entirely unknown / where a new life waits to be written / before i go / i take out everything i had hidden in my heart / and say goodbye to it / piece by piece / memory rises' — the poetic capstone to his move to Asia. Steipete's double: 'cli is nice, having UI visualizations and your team where you work is nicer' (1067 likes) and 'I added the rotation USB protocol and told my claw to look around — now it's having fun with my 360 webcam' (208 likes).

Nikunj's eSIM joy (170 likes): 'Mother in law is visiting from India… Wants to do serendipitous date night with her grand daughter. Realized she didn't sign up for roaming… I get a data eSIM in under three minutes and she's connected. What a wonderful world we live in 🤗 (real ones know this would historically be a trip to the store or calling the carrier).' Peter Yang on AI-fluent human assistants (21 likes): 'I've been working with Char, my (human!) assistant from @oceanstalent for 6 months. The best assistants today need to know how to work with AI agents. Char uses Claude Code and Codex to help me run podcast post-production, draft show notes, and make clips.' Plus top-down vs bottom-up evals from Shreya (43 likes). Madhu Guru's evals Part 7 (110 likes): the Goldilocks principle — evals should measure at the level of the jobs to be done, not just the final answer; don't build a golden set of 'right' stock picks.

A re-run in today's feed: AI & I's conversation with Microsoft CTO Kevin Scott (originally covered in our 2026-08-14 issue) — the agentic web as this year's big story: MCP playing HTTP, NL Web playing HTML; the capability overhang requiring the whole industry to close the gap; agent memory being transactional and blocking complex delegation; Conway's Law risk (you ship your org chart); agents needing identity and permission systems; permissionless innovation beating gatekeeper middlemen; the woodworking CNC debate as a mirror for programmer-agents; and the async agent future ('go get this thing done').