Thibault Sottiaux (OpenAI) announces: 'Codex 🤟 Ona. Beyond excited to work with Johannes and team to build the future.' 990 likes.

Ona is the team behind the ona_hq coding agent infrastructure. This acquisition signals OpenAI is investing deeply in the scaffolding layer around Codex, not just the model.

Sam Altman confirms the Ona acquisition: 'Really looking forward to working together!' 2166 likes—the highest engagement tweet of the day.

swyx adds context: 'Congrats to our friends @ona_hq on joining @openai! See their talk here for alpha on what's next for Codex.' 27 likes.

The link to their talk suggests Ona has been working on infrastructure that directly feeds into Codex's roadmap—likely the agent orchestration and harness layer.

Amjad (Replit CEO): 'For the first time, I'm vibecoding with ZERO frustration and in a complete state of flow, so much so that I'm running out of ideas. Typically I have so much backlog of things I want to add, but after Fable landed on Replit, I'm almost certain I don't need more IQ for vibecoding, just cheaper and faster models, and we're done here.' 1165 likes.

This is a striking statement from someone who has been building coding tools for years—the claim that the IQ problem is essentially solved and it's now an economics problem.

Dan Shipper's contrasting experience: 'Had an idea for a big Fable project, set it up, and let it cook. Came back an hour later and it had triggered the safeguards and fell back to 4.8 10 minutes in. Back to Codex.' 420 likes.

The safeguards issue is real—Karpathy also noted they're 'a little too trigger happy for launch.' The Fable/Mythos split means the unrestricted version exists but isn't widely available yet.

Amjad on Fable economics: 'Replit Agent team did a great job making Fable cost stomachable. The lack of mistakes net net makes it more affordable.' 67 likes.

The argument: Fable is more expensive per token, but because it doesn't make mistakes that require re-runs, the total cost is lower. Quality is a cost saver.

swyx's Loopcraft manifesto: 'The entire game of the next century is to be able to stack loops as effectively as possible.' In early days, it's valuable to know when to go DOWN a loop when things go wrong (for reliability). But it's more valuable to know how to go UP a loop as models improve (for leverage). 'If you don't figure out how to do this, don't be salty when you lose to those that do.' 29 likes.

This is a framework for thinking about abstraction layers in AI-assisted development: each layer is a 'loop' of automated work. The skill is knowing when to dive deeper into a loop for debugging and when to rise up for more leverage as models improve.

Rauchg showcases Vercel + Shopify: a fully custom Next.js storefront on headless, built with v0 + Cursor. 500+ orders processed in 2 minutes. 'So long on the web. Anyone can now dream → build → ship → sell.' 588 likes.

This is the Jevons paradox in e-commerce: making it easier to build and deploy stores doesn't reduce demand—it explodes it.

Steipete shows off getting Chris (likely a non-engineer) to do a PR with Codex. 802 likes.

The 'getting non-engineers to ship code' demo is becoming a recurring pattern—each new model generation produces a viral moment of someone who 'shouldn't' be coding shipping something real.

Levie shares Box's survey of 1,640 IT leaders across the US, Japan, and Europe: the companies that adopted AI the most are planning to grow headcount the most. 'It's actually quite intuitive—the companies that become most productive want to reinvest back into the business to keep getting the gains going.' 124 likes.

The narrative of jobs being wiped out assumes companies take a fixed approach. In practice, AI causes companies to want to light up more engineering projects, sell to more customers, automate more processes. That leads to more work for people, not less.

Nikunj: 'What you work on has never been more important. Make sure it's fun. Make sure it's insanely ambitious. That's the greatest moat.' 103 likes.

In a world where everyone has access to the same AI tools, the differentiator isn't capability—it's ambition and taste in choosing what to build.

Logan Kilpatrick (Google AI Studio / Gemini API lead) talks with Training Data about the agentic era at Google. Key themes: (1) Antigravity is the new through line for Google—it's not just an IDE, it's an ecosystem (IDE, web, CLI, SDK) that powers agents across Search, Gemini app, Cloud, and AI Studio. The same harness that codes is the harness that runs agents everywhere. (2) On coding competition: 'December, the narrative was Google had won. Then the agentic coding stuff happened and the narrative shifted fast.' He acknowledges Gemini is behind on coding mindshare but says the 'avengers of AI' team is pushing hard, and 3.5 Flash surpassed every previous Pro model on coding using only post-training gains. (3) 'The model eats the harness'—what we call 'the model' today is already not just weights; it's an expanding system of tools, containers, and agent infrastructure. The scaffolding gets absorbed into the native model system over 12 months. He calls for a 'harness bench' to measure how models adapt to different harnesses. (4) Omni as a true single-model world model: not routing to different sub-models, but one model that handles text, audio, video, image editing. The video editing use case is the first to ship because it's the most mature. (5) On DeepMind culture: 'Demis is a Nobel Prize scientist, Sam is one of the world's best businessmen, Dario is esoteric—you feel that in each company's DNA.' The scientific approach at DeepMind keeps them focused on solving real human problems, not just pushing benchmark numbers. (6) Narrow superintelligence in coding is already here: 'It feels like it hasn't taken away from human developers. It's an accelerant. I feel like I have more agency.'