LukeW on design feedback loops for AI agents: system gets you 90%, markup fixes the rest
LukeW · x · 2026-11-05
Luke Wroblewski published an article on closing the gap between design systems and AI-agent-built products: no design system covers all layout/state combinations, and systems must stay lean to avoid spiraling in complexity.
- First pass is driven by the design system plus agent instructions, reaching 90% (earlier models only managed 50-60%)
- Humans then mark up mismatches—tried Figma comments, annotated screenshots, video demos—and agents fix each one until variant and composite layout issues are resolved
- Takeaway: design system + a rapid visual feedback loop keeps agent output cohesive with product design and brand
More from coding & agent
- Self-hosted AI platform: 3 old GPUs, vLLM per card, and a semantic router as one endpoint — JeremyCMorgan · 2026-11-05
- Sketch to printable STL in 4.5 minutes: an open-source Kimi K3.1 Pro CAD workflow — Arindam_1729 · 2026-11-05
- Open-source X (Twitter) MCP server lets agents post, search and manage accounts — modelcontextprotocol · 2026-11-05
- Cognition launches SWE-2 Flash, a 276B-A12B coding model that beats all US open-weight rivals — carlesgelada · 2026-11-05
- 27 coding agents, 14 days: 692k lines, 1,269 PRs, and the 3-vendor cross-review rule that worked — oliver-zehentleitner · 2026-11-05
- Product designer ships Stardew-inspired pixel town game with pure JS + Three.js — algo_diver · 2026-11-05