Sottiaux's triple: the reset (10173 likes, top of the day): 'We shipped four things that were deemed good to great and some math proofs, but the vote is clear and the community demands a reset. I did calibrate it and it *seems* that the game is rigged in reset's favor, but such are the rules at the moment. Therefore... the reset has been processed. Enjoy!'; the keep-it clause (2256): 'And we won't unship the improvements. You get to have your cake and eat it too. See you tomorrow for Day 3!'; and the good-evening beat (1309). Cherny's manifesto trio: the main (8943): 'I am surprised that people are surprised this is how I prompt Claude. Talk to Claude the way you would a coworker. There's no secret to prompting. There's no need to be overly scaffolded or prescriptive for most tasks — give Claude a goal, and it will figure it out. Back in the Sonnet 3.5 days, your prompt mattered a lot. Nowadays, it's much more important to communicate: 1. What you want it to do 2…' — plus the actual prompt drop (1119) and another example (232). Altman's awe triple: 'looking up at the stars with extra awe tonight. "thy sea is so great and my boat is so small."' (4460); 'thank you to the machines, and the structure of reality, for letting us understand a little more' (1458); and 'thank you to the untold number of people who put in the technical work, brick by brick over the generations' (1639). Swyx's poll (19): 'what is your default/workhorse coding agent today, Oct 2026?'

Thariq's trio: the abstraction thesis (4206 likes): 'working at a higher level of abstraction has always required understanding the lower level ones, I don't think coding agents changes this'; the architecture shift (633): 'we're increasingly going to be moving towards Claude's "brains" in the cloud and giving Claude "local hands" to operate on your computer. I talked about this a bit on latent space'; and the edge cases (33): 'if Claude can only access your files when your computer is online, it might be effectively blocked on doing work until your computer is back on, so perhaps you want some sort of sync but there are edge cases there.' Rauchg's double: the GTA joke (5598): 'At this rate AI will get us GTA 7 before GTA 6'; and the confidence-threshold feature (164): 'Thinking fast and s̶l̶o̶w̶ a bit less fast under a confidence threshold. A very simple feature that I think will be extremely impactful for at-scale AI decision-making.' Amjad's double: the decompilation thesis (3239): 'What's happening in the AI-powered reverse engineering and decompilation is absolutely insane. Pretty soon all software will be de facto open-source. AI is coming for everything and everyone' — plus the full conversation (170). Woodward's favorite from DevDay (92): 'Mask-based editing is my favorite from today, much more coming soon!' And Garry Tan's optimization grind (9): 'Maybe 3 more prompts got it to 35fps on 640x480 pulling from 8 CPU cores per frame.'

Claude's trio: the Every case study (1669): 'The Every team runs as much work as possible through agents. To help the whole team share skills when a new model comes out, they built a company agent on Claude Managed Agents that everyone uses in Slack. Once it caught on internally, they released it to their subscribers'; the Google handshake (947): 'In Claude, paste a Google file link or ask for a new doc, sheet, or deck, and it opens beside the chat for you and Claude to edit together. Access follows your Google sharing permissions. In beta on all paid plans'; and the conversation link (96). Shipper's trio: 'we couldn't have built the @every agent without Claude Managed Agents' (132); 'this is actually true' (311); and the Slack marketplace app (49). Steipete's team-claw hookup (82): 'I hooked up our team claw to X to trigger work faster. Unassigned sessions are for anyone to grab. Our agent looks who worked on the related code last and pings people on the server. Whole thing was a prompt and team server extended itself since plugins are now hot reloadable.' Levie's cyber thesis (164): 'AI is going to now create a whole new level of work for security dealing with the increase of vibe coded issues, agentic attacks, and even accidental agent swarms hunting for data. OpenAI + Hugging Face is just a preview of what's to come.' Turck's meme (797): 'My kids when I take care of home logistics.'

The rest: Peter Yang's trio — the Fourthwall ask (7): 'My 8-year-old and I want to use her drawings to sell t-shirts and make her first dollar online'; the Muse feature request (43): 'it'll be incredible if muse can wait on a call for me then dial me in when there's a human on the other line'; and the suburbs-wine take (19). Nan Yu's maintenance worry (25): 'I worry that this will cause extreme degradation in systems that no one loves to maintain, but someone has to.' Zara's naming lament (99). Nikunj's trio: the VC dopamine warning (13); the NYC trip (123); and the thanks note (1). Aditya's pair: the political-spam rant (14) and the Keith Madams pod (11). Plus Garry Tan's SF-politics pair (52/32).

A brand-new Unsupervised Learning (published 2026-10-06, new to this digest), hosted by Redpoint's Jacob Efron, with Josh, co-founder/CEO of Applied Compute — the 16-month-old 'new AI hyperscaler' betting that post-training wins inference. Core threads: 'owning your intelligence' is about flexibility and control, not distrust of labs ('models have been pulled from products more than once, particularly in coding'); post-training pushes the Pareto curve out — the more out-of-distribution your data, the more capability lift; the future of OOD data is online RL on production inference — 'judgment traces' that codify how each company actually decides; the blocker of continual learning is data-efficient training from sparse rewards, an unsolved problem; RL is 'a hill-climbing machine — the hardest part is defining the hill to climb, which is why evals are so important to build and safeguard. Your employees are not fungible… same thing with these models'; post-training wins inference — the largest workloads post-train first, and training and inference are 'two sides of the same coin' (train tool-call parallelization → disaggregated prefill/decode on multiple chip types; make the model 10% more token-efficient to cut the token bill); the inverted pyramid strategy (training at the bottom — few can do it; inference; routing; harness/context at the top); the Harvey review-table case study (task + environment + verifier); Satya's 'as many models as firms' — 'great models set the floor, how you optimize them sets the ceiling'; quickfire — RL generalizes only to adjacent jagged points ('train on math, transfer to legal is not really there'); what changed his mind: Jevons paradox is real ('every price cut, usage just spikes… cost optimization drives over capabilities'); Jev as enabler not replacement (billion-token scale classification on rollouts, massive trace analysis); 'the best engineers I know all learned to code before AI' — don't delegate your thinking, learn to explain your code ('if they say Claude did it, we're not excited'); he's 'short' the data vendors (Merkor/Surge) — not the companies, the paradigm ('data is becoming synthetic; paradigms change — SFT → RLHF → RLVR → robotics; the skill is pivoting'); the alpha that permeates the enterprise boundary ('when bankers leave, they don't take the proprietary data'); banks as 'a thousand layer cake'; and reward hacking stories (package exploits for easy reward; cyber as 'the perfect optimization problem').