ai-lab.md · build: passing ✓ · authored by kunlin × claude
Exploring Agentic AI in Product Design
Field notes from three scales: an AI agent I designed at work, the workflow I proposed so our product team could build AI features, and the tools I vibe-code after hours. Same method each time: a living spec, an AI that executes, a designer who owns the judgment.
deliverable
craft
handoff
the edge
AI didn't shrink the design job. It moved it upstream — from drawing every screen to defining what "right" means.
// the loop · one method, three scales
⟲ the last step feeds the first — the spec never goes stale
01 / product scale
NIA: designing an AI agent inside a $1.4B B2B platform
🚀 Our first AI feature: a shopping agent that helps B2B buyers compare products, get answers, and check out without leaving the page. Usability testing moved it from a full-page modal to a persistent sidebar, so buyers never lose their place.
fig.01 · nia, persistent sidebar agent · shipped to production
02 / team scale
Living Spec: how our product team learned to build AI features
🚀 When we started building AI features, no team knew how. I proposed one living spec that holds everything, where every team claims its part: product, design, engineering, data, QA, and security. AI drafts the first pass of each section, and a named person signs it off. I designed it as a UX designer who doesn't write code, and piloted it on Nia.
// ai-feature-spec.md · every section has an ownernew for AIquestions regular features never had
// workflow · two inputs, one living spec, five phases
// what made it work
The spec never went stale. Docs usually die after kickoff, but this one ate every decision we made and briefed the AI better each week. The most useful thing I handed off was the document, not the screens.
03 / personal scale
Vibe Coding — I don't write code. I write the rules AI follows.
🚀 Two working tools: a Claude skill that tailors my resume without breaking a pixel-locked layout, and a live job-search dashboard in Next.js. Same loop as the enterprise one — plus a harness: automated checks every build must pass before I even look.
Job Tracker ● Live
127
applications tracked · Next.js
fig.02 — one build cycle: red, patch the rule, green, ship
// workflow — two loops, and the harness only grows
// the unlock
Two loops. The harness catches everything measurable — and it only grows: each failure I catch by eye becomes a new automated check. The agent I work with next month is strictly better than this month's. Continuous learning, designed — not hoped for.