Self-initiated · AI-native

AI Lab.

Client work is where I ship. This is where I find out what AI-native actually feels like: a field guide of trust patterns you can click through, a multi-agent tool that audits real websites, and product concepts built as working prototypes rather than slides.

Everything here is labelled for what it is. Shipped work says shipped, prototypes say prototype, and there are no invented numbers on any of these pages.

12 patterns · every one a live demo

Twelve ways to make AI feel trustworthy, not just impressive.

The interface patterns I keep reaching for when an AI has to earn a person's trust: confidence with receipts, intent preview, the autonomy dial, an audit trail you can read. Each one runs live on the page, and each says where I actually built it.

Open the Field Guide

Working tool

Point a swarm of agents at any website.

Specialist agents fan out in parallel, adversarially verify every finding, then synthesize one prioritized report. I designed the orchestration, the verification pass and the human-in-the-loop review, and built it.

audit · acme.com12 confirmed

AccessibilityContent & voiceVisual hierarchyIA & navPerformanceMobile

verified · 1 report

Try the live demo

Concept · operable

FinVista, Agent Era.

I designed a lending desk where executives run the journey by hand. This asks the obvious next question: what does that desk look like when agents do the processing and the executive's job becomes judgment? Run a morning shift yourself, including the escalation you are forced to handle.

The shipped numbers are real. The agent era is a concept, and no bank has deployed it.

Operate the Loan Desk

Product concepts

Designed end to end, then built to click through.

Both are self-initiated and reconstructed from secondary research. No users were tested and no outcomes are claimed. What they show is how I think an AI-native product should behave when a person still has to stay accountable for the decision.

live prototype

Slate

The AI sources candidates, ranks them, and drafts the outreach. The recruiter stays the one who decides. Designed end to end and built as a working prototype.

Read the case study

live prototype

Crux

Five AI agents run a bank's operations while a human keeps every call. The five agent-trust patterns, built as a working prototype.

Read the case study

Built with

Claude Code · Claude · ChatGPT · Codex · VS Code · GitHub · Figma

The prototypes are real front-ends, not clickable mockups, so the interaction details I care about are things you can try rather than things I describe. I direct the work and review every output; the AI does the typing.