DESIGN LAB · AI + DESIGN SYSTEMS · 2026 · CURRENT FOCUS
AI builds whatever your system already is. I built CHIP to see it first.
AI doesn't fix a neglected design system. It sends you the bill.
01 · The judgment layer
The agent never moves silently
The human stays in the judgment layer: watch, catch, draft, approve, log.
AI industrialises whatever is already there. An agent reads what you wrote, not what you meant, so a human-fine design system can be full of silent drift the moment an agent consumes it: hardcoded values, vague layer names, thin docs.
I kept learning how to make systems agent-ready but had no way to see where my own systems were drifting before an agent amplified it.
Then I wired the readiness map into CHIP, the OS I built for the hackathon. The governing decision: it can draft, but it never acts without me.
The approve inbox. The agent drafts, the human approves, every action logs. April hackathon build, BELLA parchment skin.
02 · My own systems
Run it on my own systems, not a client's
NDA-clean and more honest.
I finished the course chapter that treats a design system like a car: a check-engine light and a ten-station inspection.
Instead of highlighting and moving on, I pointed the inspection at my own systems: design system, portfolio, content engine, and scored each on agent-readiness.
03 · In public
Build it in public, labelled a prototype
Honesty beats polish.
CHIP shows both what I'm thinking about and where my own systems are drifting, in one place. Building in public, turning red cells green out loud.
Shipping is the proof. No invented numbers.
Constraints: Five days, solo, for the Anthropic Claude Code hackathon. An honest prototype, labelled as one.
The friction log, kept in public. April hackathon build.
04 · The machine
The hard part isn't the tokens. It's keeping a human in control while the machine moves fast.
That is the work I want to do with a team: the design-system rigor, plus the guardrails that let AI move fast without losing the human call.
I don't want to just take notes on what I learn. I want it to change how my systems run. CHIP shows both what I'm thinking about and where my own systems are drifting, in one place.
Credit: Brad Frost, TJ Pitre, Ian Frost (Southleft), whose AI and design systems course seeded the inspection thinking. Evidence uses illustrative data and BELLA tokens; nothing from any client.
The system map, pointed at my own work. April hackathon build.