Overview
Today felt like a tug of war between bigger models, quicker workflows, and the human messiness wrapped around it all. OpenAI pushed GPT-5.4 and a new Codex speed mode, while a wave of tools promised to make building, designing, and creating feel more hands-on. In the background: creators comparing platform payouts, investors backing industrial software, and AI leaders getting pulled into politics and public apologies.
The big picture
The pace is the story. Models are racing towards longer context and more autonomous “computer use”, but the more interesting thread is what people do with it: smaller, sharper tools that help you point an agent at the right part of the UI, keep design systems tidy, or iterate on creative work without losing control. At the same time, trust and governance are still wobbly, whether that’s users protesting model retirements or executives walking back hot takes under political pressure.
GPT-5.4 lands with computer use and a 1M context window
@OpenAIDevs announced GPT-5.4 with native computer-use capabilities, a huge context window in Codex and the API, and a push towards longer, tool-heavy workflows. The pitch is clear: fewer brittle handoffs, more work done end-to-end, and stronger agentic coding when things get complicated.
https://x.com/OpenAIDevs/status/2029620984853188738
Codex gets a “/fast” mode, and the replies are not gentle
Sam Altman’s quick note about “/fast” should have been an easy win, but it dropped into an already tense mood. The replies read like a referendum on model deprecations and how much people dislike having a favourite tool pulled out from under them.
https://x.com/sama/status/2029623948980416681
React Grab makes UI edits feel like pointing and asking
@aidenybai shipped React Grab, an open-source tool that lets you select an element on a page and then tell Claude Code or Codex what to change. It’s the sort of small idea that can save hours, because it feeds the model the right context instead of hoping it guesses which button you meant.
https://x.com/aidenybai/status/2029603067927351506
Figma tries to end “detach hell” with Slots
Figma’s new Slots feature is a design-systems quality-of-life fix: customise component instances without detaching them. If you have ever watched a library slowly rot because everyone detached “just this once”, you already get why people are cheering.
https://x.com/figma/status/2029602684781965489
A prompting trick: ask for a directory of landing pages, not a single page
@learn2vibe shared a simple hack that matches how humans actually choose designs: give me options I can compare. Instead of generating one landing page, the prompt asks for a little directory of multiple pages, so you can click around, react, and iterate with clearer taste and less back-and-forth.
https://x.com/learn2vibe/status/2029602866701422895
Luma Agents bets on co-creation, not solo automation
@LumaLabsAI introduced Luma Agents, positioned as creative partners that “see what you see” and build alongside you. The subtext is about pace and continuity across assets, keeping a project’s intent intact while you explore variations without starting from scratch each time.
https://x.com/LumaLabsAI/status/2029648371766423958
LTX-2.3 claims fast 4K video generation with native dialogue
@LTXStudio rolled out LTX-2.3, calling it their most production-ready model yet. The headline features are speed, 4K output, and built-in dialogue, aimed at reducing the usual patchwork of separate tools and fixes that AI video still tends to require.
https://x.com/LTXStudio/status/2029655155847921875
A free “super agent” from China lights up GitHub talk
@JulianGoldieSEO highlighted DeerFlow 2.0, pitched as an open-source agent framework that can research, write code, create files, and deploy. The excitement is obvious, but so are the questions people keep asking about reliability, security, and what happens when you let an agent run loose on real projects.
https://x.com/JulianGoldieSEO/status/2029602845545144357
Grok voice mode adds attachments, as the app’s momentum grows
@grok showed off voice mode that can take attachments and images mid-conversation, a practical upgrade for anyone who wants hands-free help with documents. The broader backdrop is that Grok’s app usage looks lively, with the platform trying to turn “chat” into something closer to daily utility.
https://x.com/grok/status/2029608542513361318
AI, politics, and public apologies collide at Anthropic
@ns123abc shared a clip of Anthropic CEO Dario Amodei apologising for an internal memo that became public and politically charged. It’s a reminder that the AI race is not just technical, it’s also about relationships with governments, procurement, and how quickly a stray message can become a headline.
