We tested Figma's new AI agent on a real page for an actual Settings screen, not a demo. It reads as literally an assistant: clean file, clear prompt, and it helps with UX enhancements and autolayout.

We didn't accept the output as-is. Some of it broke. So we audited what worked, what didn't & why AI didn't replace the design process, it just moved where our attention needed to go!

Really like this take. AI is definitely becoming a great tool for speeding things up, but the design thinking and review part still matters a lot.

DONT MISS THIS ONE: I've spent the last few weeks building something I've wanted to finish for a while and add new skills to my pipeline while working on the project: a cinematic, terminal/HUD-style interface for a fictional tactical operations system, entirely in the browser. (I believe this might be a really good solution to run some type of motion projects during filmmaking time directly on the visible probs) Here are some details: Behind a simulated boot sequence (fake network handshakes, IP auth, animated logo reveal) sit four consoles: → STATICPEN: a satellite recon view with live target tracking that shifts from optical to thermal scan on inspection → SILENTPEN: a ground tactical console with four switchable sensor modes: Direct, Lidar, Heat Scan, Ghost → SEERPEN: the deep one. A real 3D globe (Cesium) rendering 26 live/static intelligence layers, conflict zones, ship traffic, flight tracking, GPS jamming, orbital satellites, paired with a live news feed and a local-LLM "analyst" that reads the map context and can act on it through chat → STRATUM: locked for now. Select it, and the system denies you clearance and logs the attempt. A dead end on purpose. Design-wise, the goal was to make it feel like a live system, not a mockup; every terminal readout, scan line, and boot log is actually animating in real time, not a static screenshot dressed up to look busy. Designed fully in Figma, first run and perfected on a local server and moved to GitHub with some changes to able to fit the webpage format. It was built with React 19, TypeScript, Vite, Tailwind v4, Cesium, and a small Express backend aggregating real news and live layer data. While I've FE skills, they were limited in most cases; I've tried a new MCP server system inside Figma, and with VSC and Claude, I was able to pull out my design, put more detailed meaning to it and run it fully with interactions. While I'm against the ML systems to some degree, my approach is to use that as a tool to solve my technical problems. And my design project became a fully functional project. I believe designers will evolve their mindset to create more valuable products with more extensive tool options. Let me know what do you think =) Full case study on Behance: https://lnkd.in/dNd_PRqu Live demo: https://lnkd.in/d2cp9mcN

Amazing! Great Work! 👍