All experiments

Level 1I decide, AI buildsProduct and UX calls are mine. AI is the production tool.
MarkeTrade markets feed
Web app

MarkeTrade

Claude Code and Figma MCP to build; every product decision mine.

No projects for this platform here yet.
Level 2Agents build, I directA pipeline of agent roles builds. I steer and override.
BiltOn foreman app home screen
Mobile app

BiltOn Permit-to-Work

Five-role agent pipeline; I overrode it where field context demanded.

No projects for this platform here yet.
Level 3AI runs the processFrom goal, guidelines and benchmarks to a working app.
Cust2Mate smart cart screen
Tablet app

Cust2Mate Smart Cart

An AI-driven, end-to-end product creation process.

No projects for this platform here yet.
Level 4Agent decidesOnly a high-level goal. Research, UX, design and build are the agent’s.
MarkeTrade Agentic markets table
Web app · Compare with Level 1

MarkeTrade Agentic

Same product as MarkeTrade. Same brief, no designer.

No projects for this platform here yet.

DESIGNING FOR AGENTS · BUILT AT LEVEL 2

Northfield – Fraud Review Console

An agent stops risky payments and a human makes the call. I directed Claude from a dashboard design reference, then polished it with a design-polish skill.

Northfield fraud review queue

What I’ve found so far

Agents reach a working first version much faster. Three things still need a designer.

01

Polish

A first version only becomes a product through review and iteration.

02

UX judgment

Agents’ UX calls need pushback, like cards with no visible Yes/No action.

03

Taste

Left alone, agents design generic. Direction and judgment make it distinctive.

Privacy Preference Center