Figma Links Aren't Enough: What Context to Give AI Coding Agents Before They Touch UI
T07NAD0 · reddit · 2026-09-16
A developer shares observations from building UI with AI coding agents: a Figma link or screenshots alone are nowhere near enough. The gap between design and build shows up as drifting spacing, softened hierarchy, slightly off typography — the UI works but loses character.
What works better is giving the agent clearer context before it starts: not cleverer prompting, but writing down the decisions that should stay fixed so the agent isn't inventing them. He doesn't expect the agent to pick the design direction — he uses it for execution and repetitive work, keeping direction, references and rules himself.
His take: most people aren't overcomplicating AI coding, they're doing the opposite — one prompt like "build me an app that does X" while expecting the agent to understand decisions never shared. He asks what others treat as the source of truth: Figma, written context, the codebase, or a combination.
More from coding & agent
- When an LLM plans an executable agent DAG, where do you draw the trust boundary? — Repulsive_Sugar_5252 · 2026-09-16
- "Never again": user rants Nemotron 3 Ultra can't find .github and stalls for 90 minutes — dwerps · 2026-09-16
- Personal AI's real test: does your judgment survive after you close the window? — sujingshen · 2026-09-16
- Coding agent so fears AWS costs it wrote a test banning NAT gateways — mimi10v3 · 2026-09-16
- Muse Code raises contributor tier limits and resets usage for all subscribers — giffmana · 2026-09-16
- OpenAI reportedly orchestrates 10k internal agents without quality loss, while users' sub-agents 'just burn tokens' — RexDouglass · 2026-09-16