Daniel Chu / AI-native builder

I build agent systems that survive real work.

I lead Copilot Mobile product work and build the systems myself: multi-agent development loops, eval gates, workflow replay, and recovery paths for product teams using AI in production.

This is my public lab notebook for what breaks when agents touch files, tools, memory, side effects, and decisions that need review.

See selected work Read the notes About me

Shipped Production multi-agent development system for Copilot Mobile

Builds Hands-on agent workflows, eval harnesses, and recovery tooling

Studies Atlas experiments in tool use, memory, automation, and safety boundaries

A warm studio desk with AI workflow sketches, interface cards, and evaluation notes. — **The workflow is the test.** Agent loops, product constraints, review paths, and repair in one frame.

Start here

Product Development Was Never Just About Coding

Agentic development makes software easier to build, but product work still depends on customer understanding, trust, onboarding, and human adoption.

Read the latest note

Builder evidence

Systems, not slideware.

Shipped system Agentic development

Copilot Mobile Multi-Agent Development System

A production multi-agent system that turns product intent into mobile code changes with evals, review, and recovery built into the path.

Personal lab AI reliability

Atlas

My private AI workflow system for testing tool calls, context, memory, scheduled jobs, sensitive data boundaries, and agent reliability.

Contact

If you are building frontier AI systems that need product judgment and hands-on builders, I would be happy to compare notes.

LinkedIn X