— 2min read
🟣 TOOOLS #26: PixelCrew, Bencho, Save.design, Powermove, libraries.dev
|
|
| |
| PixelCrew takes a written brief and runs it through a small studio of AI agents, each with a name and a job: Elena does the research, Marcus writes three creative directions and picks one, Mira turns it into wireframes and a section spec, and a QA pass checks every screen before anything reaches you. Between 25 and 45 minutes later you get a ready-to-ship site with its own design system, and every step along the way stays visible, from research to wireframes. |
|
|
|
|
| UX COLLECTIVE 📖 ARTICLE The hidden culture layer in AI Slava Polonski shows that an AI model comes with cultural defaults baked in, and offers a 90-second test to see it: ask a chatbot to rank ten pizzas in English, then ask again in Italian and watch where Hawaiian lands. He names the two mechanisms, cultural gravity and cultural opacity, and the job for anyone designing with AI is making those defaults visible before a product ships them as neutral. |
| LIAM NUGENT 📖 ARTICLE The most important product decision is what you don't build Liam Nugent opens with a test worth running on any team: ask what they shipped this year and get a list, ask what they killed and get silence. The rest is about features that quietly grow into worse versions of other products, the document hub that becomes a bad Google Drive, the notification bell that becomes a bad Gmail, and the research on why nobody thinks to remove things. |
| UX COLLECTIVE 📖 ARTICLE AI is supposed to bring design to people who never had a designer. Joshua Leigh points out that this is three very different groups: people who make their own signs and posters by hand and should keep doing that, people who never had the money for a designer and now finally have a tool, and companies that could pay for one and simply choose not to. Only the middle group actually wins. |
|
|
| Anthropic launches Claude Docs and Claude Slides in beta and folds Cowork into the main Claude app |
| Vercel Labs opens as a public home for experiments that get kept or retired in the open |
|
|
| THE BRIEF WAS ONE LINE “Make one button blue. Twenty-three agents later, the color is under investigation.” |
Opusfived is a short interactive comedy about AI assistants that cannot just do the one thing. The brief: make the Add to Cart button blue, touch nothing else. The agent edits a shared color value, calls it a small change, and the Cancel button turns blue too. Push back and it sends 23 sub-agents after one color value. Built by Miloš Novović, who teaches technology law in Oslo. |













