A design pipeline, run by agents.
I built a multi-agent system that runs the full UI/UX loop in Figma — inspiration, a real design system, screen building, and a scored audit — with me approving every stage and a hard quality bar on the output. Not vibes. A system.
Four stages, from direction to scored screens.
One agent per stage, each with a single, narrow job — and each one hands its output to the next. I stay the director; the agents do the throughput.
Set the direction
Reads references — images, Figma, notes — into a written visual direction.
Build the system
3-tier tokens, 30+ components with full states, grid, icons — in Figma.
Wireframe → hi-fi
Builds each screen from the system only — low-fi first, then high-fidelity.
Score the work
Grades fidelity, direction-match & UX. Below bar → surgical fixes, re-audit.
Four specialists, one shared system.
Each agent has one job and one contract — a clear input, a clear output. None of them improvise; every screen is built from what’s already approved.
Direction-setter
System architect
Screen builder
Quality gate
Nothing ships under 80.
Every screen is scored on three axes. If any falls below the bar, the auditor writes precise fixes and sends only those back to the builder — then re-scores.
Weeks of work, in days.
Same scope on both sides, from direction to QA’d hi-fi. The pipeline doesn’t cut corners; it collapses the production time and shifts the effort to direction and review.
Estimates for a representative 0→1 build, one designer — production time only. Scope and mileage vary.
Speed, without losing the craft.
0→1, faster
Direction to a scored, on-system hi-fi screen in a fraction of the usual time — because the system does the throughput.
Consistent by construction
Every screen is built from one approved design system, so nothing drifts. The system evolves only with my sign-off.
Human-in-the-loop
Approval gates at every stage and a hard quality bar mean I stay the director — the taste and the calls are still mine.