François KY 👨‍💻🛫

Code @Sanofi - #GradatimFerociter

[Cursor @cursor_ai]

We're introducing Cursor Automations to build always-on agents.

[François KY 👨‍💻🛫 @KY_F]

I am on a subway the lady next to me seems to catch with work before arriving at the office. It has been 8 minutes since we left the station and she hit the power on button still not able to log in. I am shock as Mac user. #HP

[OpenAI Developers @OpenAIDevs]

Introducing WebSockets in the Responses API.

Built for low-latency, long-running agents with heavy tool calls.

https://t.co/qmOAhidk7o

[Milena Surreau @MilenaSurreau]

Ça c’est moi qui me traine le cul (dans la pisse et les dechets) parce que la mairie de @VilleStNazaire n’a toujours pas daigné réparer l’ascenseur qui mène à la gare depuis 10 jours.
@DavidSamzun c’est normal ?

[Chris Tate @ctatedev]

Introducing json-render

AI-generated UI. Deterministic output.

1. Define your component catalog
2. AI steams JSON
3. Render interactive UI

Let users prompt dashboards, widgets and apps - safely constrained to components and actions you define

[Greg Brockman @gdb]

Introducing ChatGPT Health: https://t.co/HnzAoH6UtV

[Josef Bender @josefbender_]

TIL about ListFormat.

A truly powerful @JavaScript function for formating lists in a natural way.

👇Here are a few examples:

[Nico Albanese @nicoalbanese10]

Lots of hype for programmatic tool calling and for good reason.

It's very cool and definitely worth experimenting with (Anthropic reports avg. 37% fewer tokens used).

@aayushkapoor_ cooked!

[Aiden Bai @aidenybai]

why does cursor's bugbot not find all issues on the first try?

i keep having to bugbot run → fix in cursor → bugbot run → fix in cursor → bugbot run

over and over again

[Lars Grammel @lgrammel]

The generateImage function supports editing existing images now.

[Thariq @trq212]

Claude Code can now control your browser for testing, getting data, etc. using our chrome extension!

[Nic Cruz Patane @niccruzpatane]

This is the new SpaceX ISS Docking Simulator game added in Tesla’s 2025 Holiday Update.

Its the actual interface used by NASA Astronauts to manually pilot the SpaceX Dragon 2 vehicle to the International Space Station.

[Matt Pocock @mattpocockuk]

Compare unfavorably to rival library and ye shall receive

[Adam Wathan @adamwathan]

It's wild that Tailwind is so ubiquitous at this point that a company can launch a visual Tailwind editor without it even seeming important to say the word "Tailwind" 😄

Proud of that honestly, and very cool demo! Cursor is a fantastic product and an exciting team to follow.

[Infos Françaises @InfosFrancaises]

Made in France : un publicité de Noël de l’enseigne Intermarché devient virale et fait le tour du monde ces dernières heures.
Un conte réalisé sans intelligence artificielle, réalisé par le studio d’animation montpelliérain Illogic Studios.

[Alem Tuzlak 🇧🇦 @AlemTuzlak]

Now that I'm awake, should I start showing you some cool things the @tan_stack AI can do?

[Matt Pocock @mattpocockuk]

Initial thoughts about TanStack AI, and its positioning vs Vercel's AI SDK:

Framing

I like that they are pushing 'no vendor lock-in' in their blog post (that's how all AI libs should be).

But I think they're subtly benefiting from the impression that all Vercel libraries (i.e. the AI SDK) MUST be trying to lock you in somehow.

The AI SDK is vendor-agnostic, with a great DX whether you use Vercel's built-in gateway or the model providers. IMO, the type safety and isomorphic features are more interesting.

Devtools (TanStack wins)

TanStack AI immediately ships with dev tools. Hell yes, why the AI SDK haven't done this already boggles the mind.

Per-model Type Safety (Inconclusive)

One weakness of the AI SDK is that the options you pass to the provider are not strongly typed. You can attempt to use structured outputs on a model that doesn't support them, and it won't warn you at compile time. TanStack AI _will_ give you these warnings.

This is a nice-to-have in my book, I usually know the capabilities of the models I'm using before I go to code.

One other common complaint is that once you're _in_ the AI SDK ecosystem, it's pretty hard to do advanced provider-specific shenanigans. No idea if this will be easier in TanStack AI - my bet is it's probably a wash.

Features (AI SDK)

I really like that TanStack AI ships with tool approvals built in. The isomorphic tool definitions are a nice upgrade on the AI SDK's - you can still have isomorphic tool schemas in the AI SDK, but not the entire tool. I imagine that TanStack's isomorphic API will be really nice to use, just like query/router etc.

From what I'm reading in the docs, the AI SDK gives you a ton more control over the stream (with custom data parts) than TanStack AI does. TanStack AI's docs are more skewed towards agents than workflows.

However, I'm intrigued to see how this develops - if anyone can create a beautiful type-safe API for streaming custom data over the wire, it's the TanStack query team.

Finally, the pure number of providers from the AI SDK is way beyond what TanStack can offer right now.

Conclusion

Don't adopt just yet. It's got a long way to go to catch up to the AI SDK's feature set. Duh, it's an alpha.

But it's an impressive launch. If they can nail the type safe streaming DX better than the AI SDK (I do think there's headroom here) it'll be hard to top.

[Anthropic @AnthropicAI]

New Anthropic research: Estimating AI productivity gains from Claude conversations.

The Anthropic Economic Index tells us where Claude is used, and for which tasks. But it doesn’t tell us how useful Claude is. How much time does it save?

[François KY 👨‍💻🛫 @KY_F]

@elonmuskADO 56

[Gavin Ching @gching]

Crazy. @OpenAIDevs just shipped ChatGPT Apps SDK UI components🔥😭

https://t.co/KQgLokjVq0

Making my life so easier to cook🧑‍🍳 will use for my next day for ChatGPT Apps

Im loving this @saitonian @kagigz and team🫰

Thank you LMDTFY 🤭