So I guess Codex has a little robot now?
(It's not as cute as the Claw'd crab)
GIF
English
Simon Willison
61.1K posts
Simon Willison
@simonw
Creator @datasetteproj, co-creator Django. PSF board. Hangs out with @natbat. He/Him. Mastodon: https://t.co/t0MrmnJW0K Bsky: https://t.co/OnWIyhX4CH
San Francisco, CA Katılım Kasım 2006
5.7K Takip Edilen197K Takipçiler
@emollick I'm fascinated by how different the "Chat" and "Work" modes in the mobile app are - I'm still trying to nail down all the differences but they are material
On desktop "Codex" v "Work" is purely cosmetic
English
If you are use the Claude everything app, you pick between Home and Code. If you pick Home you get to pick between Chat & Cowork
If you use the OpenAI everything app, you pick between ChatGPT Work & Codex. Chat is in a side menu.
Both are different on their websites
intuitive!
English
New TIL: Using uvx in GitHub Actions in a cache-friendly way
I finally found a recipe that I like for running `uvx tool-name` in GitHub Actions without downloading a fresh copy of the package every time til.simonwillison.net/github-actions…
English
More on how my HTML+JS UI for this works on my blog - I used Datasette Apps, a new plugin for Datasette that lets it host custom HTML+JS apps that can Execute SQL queries against SQLite simonwillison.net/2026/Jul/13/do…
English
This is really cool! It renders frames using a ray tracer implemented as a SQLite recursive CTE
Since state is stored in a SQLite database on-disk you can have other tools query it while you are playing - I rolled this custom HTML+JS view that adds a minimap
Peter Gostev@petergostev
GPT-5.6-Sol-Ultra built a Doom-like game in SQL: the game is running in the terminal on the left, while the SQL powering the game in the terminal on the right. How it works: DOOMQL is a game engine implemented in 2,000+ lines of SQL. Every frame, SQL raycasts the scene, calculates every RGB pixel, and encodes those pixels as coloured terminal characters. The same SQL then handles controls, movement, collision, enemy AI and combat, while Python connects SQLite to the keyboard, clock and terminal.
English
It's been about six months since OpenClaw burst onto the scene - are you still using yours? Did it become a daily driver? Any interesting lessons or anecdotes you can share?
English
(Meanwhile OpenAI's new Sol, Terra, Luna naming convention makes perfect sense to anyone who's familiar with the Latin names for bodies in our solar system)
English
I've seen a few people predicting that Opus 5 will be out soon and will be better than Fable 5, but have Anthropic clarified how their relative naming scheme works yet?
I assumed it was Haiku < Sonnet < Opus < Fable < Mythos - but is Fable meant to go between Sonnet and Opus?
English
@MaximePeabody Yeah that's a good point, Fable and Mythos are meant to be equivalent strength (and effectively identical) models but Mythos isn't restricted by the extra guardrails
English
@simonw I thought fable was equivalent to mythos, but just with the added guardrails?
English
It's annoying that you can't paste a link to a (shared) Claude transcript into a Claude Code session, because Anthropic's anti-scraping measure prevent its own tools from accessing the output of its other tools
English
@pvncher @MarcosHernanz It's confusing how Work vs Codex in the desktop app is just cosmetic, but Work vs Chat in the iPhone app is radically different
English
@MarcosHernanz Its not just for general users. The dropdown allows us to build two product views, one for devs and one for general users. There’s a lot of overlap but still many distinctions.
For instance, check the new pull request tab that is only for codex mode.
English
Codex app is becoming an app for general users.
And the TUI sucks.
Who's building an alternative?
Tibo@thsottiaux
If you were using Codex before, two things to do 1. Keep using Codex. It is for you 2. Tell your friends and family about ChatGPT Work who you always wanted to show Codex but they’re not as technical and don’t work with code every day and tell me how it goes
English
@TimTeaFan I think one fundamental issue here is accountability: it makes no sense to try and hold a bunch of matrix arithmetic accountable for something
If you're hiring employees you need to be able to hold them accountable for their actions
English
I would like to disagree, especially bc what we see where model capabilities are going and what agents like open claw and Hermes can do. it might not be much now, but it will (and does) surpass any spreadsheet. and on the other side you basically have intelligence degrading on the human side idk for the US but in Germany it is not easy to hire capable ppl.
English
The idea of "AI employees" feels so short-sighted to me - both disrespectful to humans and a complete misunderstanding of what these tools can do and how to best put them to work
You may as well start adding Excel spreadsheets to your org chart
English
@_xjdr would strongly encourage *not* running in yolo mode and instead run it with auto review instead
learn.chatgpt.com/docs/sandboxin…
English
my biggest take away from gpt5.6 so far (and fable to some extent as well) is that you only get a fraction of the value without running in --yolo mode (subagents that require approvals aren't any more useful than just running the prompts yourself) and as a staunch --yolo mode denier (legacy review every diff unc), i am finding this dichotomy difficult to cope with . i knew this was coming, and i've been preparing for it (its like 60% of the long term thesis behind code.noumena.com) but i didn't think it was already here.
English
@emollick I had an argument with it about this where it appeared to be taking the instructions around quoting from web search results and interpreting them as applying to all other sources (like documents I uploaded to it) as well
English
@emollick Fable 5 is particularly strict about not quoting more than 15 words from sources, which is absolutely infuriating when you're doing research and want it to quote the original sources to help support claims that it makes in its responses
English
I've been going on about how ChatGPT Work (and Cowork) are missed opportunities for knowledge workers, and to illustrate that take a look at Google's NotebookLM answering the same question as ChatGPT Work, with the same 70+ files.
It centers process & sources, not just outputs.
English
More interesting is the difference between "Chat" and "Work" modes in the ChatGPT mobile app
It looks like Work mode can run code that talks to the Internet!
I just tried having both modes use yt-dlp to extract subtitles from a YouTube video - it failed in Chat, worked in Work
English
Looks like the line from OpenAI insiders is the at it's purely a UI thing, not a difference in capabilities
Andrew Ambrosino@ajambrosino
@simonw capabilities and histories are the same— it’s a UI preference
English
Anyone know if ChatGPT Codex (in the new ChatGPT desktop app) is a strict superset of ChatGPT Work?
Liked if you're a software engineer who isn't intimidated by Git features is there any reason you'd ever want to switch to ChatGPT Work Mode?
English
Not a superset. They’re distinct modes for different loops: Work is for research and multi-step deliverables like decks or sites, while Codex remains the agent for repo-level coding, terminal access, and diffs.
If you’re a dev, you’ll stay in Codex. Work is just a different toolset in the same app.
English
@corbtt So sign into Facebook marketplace from the AI browser, don't give it access to your entire online life just to enable that one usecase
English
@simonw I'd put money on this being resolved in the other direction (the browser you use and the browser your AI agents use will tend to converge).
It's too useful for my agent to be able to eg. manage a Facebook Marketplace listing for me. It needs my accounts.
English
With Atlas being retired in favor of the browser embedded in the ChatGPT app I wonder if the whole category of AI-enhanced browsers is coming to a close
The security/privacy issues remain unsolvable IMO - I want my AI to use its own separate browser and stay out of the one I use
English