It's Monday. Here's what changed in AI this week, and what it means for your design work:

I read the Astra announcement on Thursday night and did not sleep much after it.

Nothing is wrong. This has been an incredible week. I moved into a real rhythm in Cyprus where the mornings belong to content, one video a day, scripted and prepped before lunch. On Sunday I record the whole week in one go and drop the files in the Drive for my editing team, and the afternoons are mine for the products.

That afternoon time has been going into a single question for weeks, which is what Limora and Stage look like two years out. I had most of an answer. Then OpenAI put out a model that can do anything you can do on a computer, and the answer stopped making sense. If the agent operates the software, the interface is not built for a person to touch anymore. The person directs by voice, and what they see on screen becomes a different surface from the one the agent works in.

I knew it was coming. I did not think Jarvis was coming this fast. More on Astra below, and I would watch the launch video before you read my take.

This is the most important launch of the year for designers, and almost nobody is covering it that way.

OpenAI shipped GPT-6 Astra on Thursday. The headline from the launch is one sentence: anything you can do on a computer, Astra can do for you. It has a 1.05 million token context window, it uses a computer natively, and it reaches your tools over MCP. In the launch demos it modelled a house in Blender and turned it into a walkable Unreal Engine 5 scene, and it ran a full design exploration for a matcha shop from vector all the way to prototype.

The 3D demos got the applause, but the thing that actually stopped me was Marc Schwaibold's clip of Astra generating UI.

Look at the spacing and the type. That is not the flat purple-gradient output we have all been laughing at for two years. The model is starting to understand what good design is, and that is a real signal about how fast this is moving.

I'll be honest with you, that is a strange thing to sit with as a designer. Taste was supposed to be the last thing on our side of the line. But the same shift is the opening: when generation becomes free and decent quality becomes common, the work moves to direction, judgement, and knowing why one version is better than the other nine. That has always been the part clients were actually paying for. We are just going to be doing it out loud now.

Do not start by asking it to make you a screen. Give it a real task with real constraints, a brand, a reference, and a goal, then watch how far it gets before you need to take over. That gap is the most useful thing you will learn this month.

Three prompts from this week, all built on the same idea. You feed in one ordinary photo and get back something that looks designed rather than generated.

You change a colour token in Figma, then you export it, then you message the dev, then the dev pastes it in, then somebody forgets the dark mode value. Every design system I have ever touched leaks somewhere in that chain.

GitFig closes it. It is a Figma plugin that syncs your variables, colour styles, and text styles straight to a GitHub repo, both directions. It speaks W3C Design Tokens, Style Dictionary, and Tokens Studio, and it keeps multi-mode collections in sync so your light and dark values stay tied together with the aliases intact. You stage changes in the panel, commit when you are ready, and open a branch or a pull request without leaving Figma.

Most sign up forms are dead rectangles. This one has an avatar that reacts to every word you type, and it took one repo and a few prompts.

💌 Know a designer who should be using AI smarter?
Forward them this email. Or just send them to logiaweb.net to join.

See you next Monday,

Say hi on X, Instagram or YouTube