How I work

I leverage AI coding agents in my workflow. What makes that work is experience: years of designing production architecture help me decide where agents speed things up and where a system needs firm structure and review.

That background is how I keep improving both the architecture and the workflow in a production environment.

Agents follow the same process

Agents open merge requests instead of pushing straight to production. Their changes go through the same CI, tests, and review as mine, and I approve anything that ships.

Every repo has a map

Each project keeps a knowledge graph of its own code, updated automatically by CI. Agents and people both start from that map instead of searching the codebase from scratch.

Standards live in code

Shared pipeline templates and repo standards keep projects consistent, so an agent working in one repo behaves the same way in the next.

Building the harness

I've been researching agent harnesses: a small orchestration layer that schedules agent jobs, runs longer tasks in loops, and keeps an audit trail of every run, with a web dashboard for routing work and reviewing results.

The goal is the same as any tooling I build: help teams ship reliable software.