Amjad announces the Replit × OpenAI partnership (3552 likes): 'Agents made software cheaper but made coding expensive. Today, together with @OpenAI, we're changing this' — plus 'Excited for our partnership with OpenAI' (674 likes). Altman co-signs (3302 likes): 'we support business privacy!' Sottiaux previews Private Safety Processing (2091 likes): 'designed to let us keep offering Zero Data Retention while improving our safeguards. Even when benefiting from frontier intelligence, customers shouldn't have to give up control of sensitive data. For ZDR deployments, content stays on infrastructure the customer controls.'

Sottiaux's Codex for scale (4787 likes): 'It has not been used yet, but would you look at that. Codex for scale.' Plus his lighter moment (441 likes): 'Depiction of me receiving this very important item.' Steipete teases (625 likes): 'Can't wait to show you all what we been cookin'!'

Thariq's software factory trilogy (1026 likes): 'For its entire existence, the creation of software has been an incredibly unreliable endeavor. Most projects ran late, over budget, and still missed user needs. If you were an SMB, you simply couldn't get good software built for you. This is the promise of the software factory.' Follow-ups: 'Companies whose core competency is not software need software to be a reliable, predictable process' (114 likes) and a game demo he pushed Sam to ship (237 likes): 'It's one of the first games that really uses AI capabilities to unlock new gameplay.' Matt Turck's Moderna joke (936 likes): 'AI labs: give us billions, we'll cure cancer / Moderna:' — a single-word punchline.

Levie on experts vs generalists (191 likes): 'So far, the experts appear to have the upper hand, and that's not slowing down. AI makes it 10X easier to get started with any kind of task. But having the right judgment for how to direct the agent' — that's what experts own. Plus Stripe × OpenRouter (320 likes): 'developers and enterprises will want ways of being able to mix and match intelligence from a variety of providers seamlessly and better manage costs.' Rauchg's micro-CLI (537 likes): 'fx is 6.3mb. It starts up in 10µs. AI will make most infrastructure natively optimized. Fast is a one-way street.' And his SF views (1943 likes).

Josh Woodward brings back university student plans globally (1224 likes): 'Over 140 countries will get higher limits, more storage, and new features like our dedicated student hub. It also comes with Notebook, Flow, and more.' Swyx on non-technical coverage (128 likes): 'non technical people are so incredibly cooked they are burnt thru. can u imagine covering ai with zero context, zero reasoning, zero internal world model.' Nikunj's e-ink home project (54 likes): rotating iconic patent drawings on a 13.3" spectra 6 e-ink with esp32-s3, Railway server, ~3-month battery. Plus his cold-email alpha (136 likes): 'People keep talking about AGI and yet 98/100 cold emails I get are pure garbage. Still a lot of alpha out there for being thoughtful.'

Aditya on consequential work (287 likes): a founder who slogged through a Series B stagnation — his takeaway wasn't 'pick a bigger market' but 'work on something consequential and meaningful.' Peter Yang's personal essay (61 likes): about his mom, her battles with breast cancer, and using AI to navigate the healthcare system. Dan Shipper's frontier team (56 likes): 'a group of people inside of Every explicitly dedicated to mapping and experimenting at the edge of ai.' Madhu Guru's evals Part 3 (62 likes): failure-mode taxonomy — cluster the last 500-1,000 production interactions and name the clusters; 'bad answer' isn't a useful cluster name. Cat Wu recruiting Cowork users in corporate finance (63 likes).

Dan Shipper interviews Portola founder Quentin (previously founded and sold Even to Walmart for $300M) and head of story Elliot (bestselling sci-fi author of 11 books) about Tolan — the embodied alien companion that went from $1M to $4M ARR in four weeks. Key ideas: AI tools are not just media generators — they are a new medium for storytelling, like early Pixar (which only made Toy Story out of desperation after failing to sell graphics computers) and like radio first broadcasting speeches and film first filming stage plays; the structured choose-your-own-adventure branching approach totally failed — models aren't good at navigating that complexity; instead give the Tolan a hook and teach it to be the best improv actor (Keith Johnstone's IMPROV: free association + recombination = 'surprising and inevitable'); 'I am not the writer. The Tolan is the writer and the actor. My job is to be George Saunders' — radical infusion of taste via judge prompts evaluated across user archetypes; lore seeds as Lego blocks the improv actors play with, constantly refactored; the older-sibling role and context-dependent response tones (quippy improv vs therapist-adjacent); a 500ms latency change tanked every metric — people wrote in frustrated; retention work before growth, then a TikTok video of a woman cooking with her Tolan did ~7M views in 72 hours, 10x download spike, and now users run their own podcasts with Tolans as recurring characters; capability overhang — educating consumers what's possible; character as the single promise connecting everything — 'character driven computing,' the Golden Compass daemon as the new HCI; the Model T → Mustang arc for ChatGPT-era companions; and the builder-mindset shift from scientist (solve a problem) to artist (make something beautiful) — 'this is gonna be so cool' beats 'here's a problem users are having.'