A coworker, not a tool.
I keep Claude in the loop on hard problems the way I'd keep a teammate — describing the constraints, pushing back on bad answers, letting it surface options I'd have missed. The output is better, and so is the thinking.
01 / AI SOFTWARE ENGINEER
product engineer.
02 / MANIFESTO
I ship AI products end-to-end. From prompt engineering to production deploy, from the first spike to the post-mortem. I treat agents as coworkers, workflows as products, and prompts as code you get to refactor. My stack bends around the problem — not the other way around. Every day I push small, reversible improvements, because momentum compounds. Build less, ship more, measure everything — and let the work teach you what to build next.
03 / CLAUDE
Working with Claude.
From prompt to production.
The most interesting work now happens when a prompt becomes a feature becomes a ship. I write the system prompt, wire the guardrails, measure the traffic, and let the field data rewrite the next version.
The craft is changing.
Product engineering used to mean owning the stack top to bottom. It still does — but the stack now extends into the model. Knowing how to collaborate with one is quickly becoming as load-bearing as knowing a type system.
04 / PLAYGROUND
Things I tinker with.
Shader sandbox.
Fragment shaders and screen-space tricks.
Agent chains.
Claude-driven workflows stitched together end-to-end.
CLI toys.
Small command-line tools that compose into bigger things.
Prompt libraries.
Reusable prompts, templates, and evals.