I'm once again very impressed by Codex models. gpt-5.2-codex is significantly faster than the predecessor and still way cheaper than Sonnet. I find it hard not using Codex at this point.

Just vibe coded a small tool I've been working on for the v2 of my/Hymaïa's LLM book: a Markdown editor that helps identify repetitive words in French 🇫🇷 text using lemmatization. It highlights overused words so you can vary your vocabulary. Built with CodeMirror and deployed on

You might think ChatGPT or MCP-UI Apps are all about interaction. And that's not necessarily true. Sure, interaction is the obvious part, probably the most powerful, but it might not be the most important. I personally see two major, and more relevant, benefits: - Presentation

Is it just me or @GitHubCopilot Pro+ has an impressive value? Plus, quotas are explicit.

I’ve been experimenting with building a small example app using the ChatGPT Apps SDK. It's been a lot of fun... until the moment when nothing seemed to work anymore, due to issues on ChatGPT; kind of frustrating, but also part of the reality of building with early tools. Still,

I quite like the new DeepSeek-OCR paper. It's a good OCR model (maybe a bit worse than dots), and yes data collection etc., but anyway it doesn't matter. The more interesting part for me (esp as a computer vision at heart who is temporarily masquerading as a natural language

An approach I'm using to keep the grasp on AI-generated code is to let AI do the heavy lifting, then doing trivial refactorings myself. For instance, renaming classes/variables, simplifying some implementations, removing dead code, etc. This helps me a lot in keeping the mental

Still, as a frontend engineer, I'm excited. It’s a natural UI-meets-AI on mobile, raising intriguing questions about standards and the future of the Web. This could be the start of something big, if they get it right.

It's tough to form an opinion on ChatGPT apps with the flood of AI news. Predicting their success is even harder. Many great UX ideas like App Clips and widgets never gained traction. Execution will key: discoverability, responsiveness, and state management can turn a solid

On stage we got: - purple gradient ✅ - you’re absolutely right ✅

I’ve gotten so used to trusting Codex’s output that I'm somewhat apprehensive in passing work to Sonnet 4.5. Am I the only one?

Grok Code Fast is a liability.

What's cool is that idb can dump the screen content to JSON, so no need to process the screenshot. As a result: - faster iterations - less flaky - should work on @FlutterDev and @reactnative

I wanted to see if this dead-simple, MCP-less workflow (cc @steipete) combining Claude + Facebook's old but gold idb would work. Turns out, it does.

We're not surfing on AI. Proof is, you could've met these two folks in 2019 at the very same conference. But how cool is it that, six years later, they’re back to blow our minds again with their cutting-edge AI workflows? -- @Dimillian https://t.co/fgk2LS3vbm @steipete

Have I missed something in the recent months, or have all new AI-based or agentic browsers largely underdelivered or, worse, failed to deliver at all? Thinking specifically about Dia and Opera.

Installed Beta 2 on my iPhone. The Liquid Glass effect is nice, but too present in such a small screen estate. It feels somehow “baroque”. As screens are larger, I find it much more pleasant on iPadOS and macOS.

And that’s all I know! For more resources: - WWDC videos -

📊 Telemetry No data is collected by Apple via telemetry.