How I work

Most engineers use AI tools. Fewer have rebuilt how they work around them — deliberately, with guardrails.

AI in the daily loop

Claude Code is my daily pair — not for autocomplete, but for whole units of work: a schema migration, a service refactor, a test suite. During an application redesign I wired Figma into the loop through MCP, so UI implementation started from the actual design file instead of a screenshot and guesswork. The practical effect isn't typing faster; it's taking on work that used to be a sprint conversation and finishing it inside a day.

Guardrails, not vibes

Speed without review is just deferred debugging. AI code gets read like a junior engineer's pull request — every line, with special suspicion for the parts that look most confident. Types, linting, and tests stay non-negotiable; the machine writes faster than I do, so the gates matter more, not less.

This site is the demo

This portfolio was built end to end with that workflow. Its content lives in a typed data layer an agent can safely edit, the repo carries custom automation for recurring jobs — adding a case study, syncing content when my resume changes, a pre-deploy quality gate — and a hook typechecks the project after every AI edit, so a mistake surfaces in seconds instead of at build time.

What stays human

Architecture, judgment, and accountability. Deciding what to build, what good looks like, and what ships under my name — none of that delegates. AI compresses the distance between a decision and its implementation; making the right decision is still the job.