anatomy
How the work actually got done
Not a portfolio. Each of these takes one piece of shipped work and describes the method: what parallelised, what didn't, what was thrown away, and what it cost.
3 of 3 case studies are placeholders. The written versions arrive through a separate handoff step (node scripts/site-anatomy-pull.js) and have not been written yet. The outlines below are real — the prose is not there.
They are published as stubs rather than hidden because an empty section with three visible gaps is more honest than a section that looks finished because the unfinished parts were filtered out.
Shipping games with parallel subagents
[PLACEHOLDER]A fleet of agents working the same codebase at once — what actually parallelises, and what has to stay sequential.
One-off sites, built and shipped same-day
[PLACEHOLDER]Small purpose-built sites as a repeatable unit of work rather than a project.
A knowledge graph agents actually read
[PLACEHOLDER]Notes wired as a graph with abstracts and edges, so an agent can scan the index and load only the nodes a task needs.