Fit: agent-directed builder
What the role usually needs
Someone who can take a brief, direct AI coding agents to build real software against it, and ship something that passes its own tests — with enough judgment to catch what the agent gets wrong and enough discipline to prove what it got right.
What I have done that matches
- Sadyr: first commit to production in twelve days, 63 commits on eight working days. Live since 30 June 2026.
- A game in development: 81,707 lines of gameplay C++, directed through 157 prompts in the two weeks that hold about 75% of that code, with 30,069 lines of automated tests and a playable build under playtest. Status: in development, not shipped.
- Cadence, the proofread-by-ear tool I built for my own dyslexia. Status: shipped, to an audience of one.
- The eval discipline: Sadyr's automated tests run against mocks by default, so I built a separate headless battery that pressure-tests the hosted agent with realistic conversations. 200 of 234 cases currently pass.
What I have not done
- The game isn't shipped. It's a playable build under playtest, not a released product. That's why the status words exist — this is what a still-evolving build gets receipts for, not a finished one.
- I have taken on build-lead work as a paid contract before: a RevOps rebuild for an anonymized client (the league system referenced below) was exactly this, and I built a HubSpot connector for another paying client last spring that's since led to referred work. This is something we're paid to do; the market has validated it.
- No experience yet directing a team of other people who are themselves directing agents — this has all been solo-directed. Managing people who do what I do would likely be a genuinely different challenge to figure out.
Evidence
| Claim | Receipt |
|---|---|
| Sadyr: blueprint to production in twelve days | /builds/sadyr/, /receipts/ |
| Game: 81,707 lines of gameplay C++, 157 prompts, 30,069 lines of tests | /builds/game-in-development/, /receipts/ |
| Cadence: shipped, to an audience of one | /builds/cadence/ |
| Sadyr eval: 200 of 234 cases passing | /receipts/ |
How I work
Plain-English direction, not code review of AI output after the fact — Claude Code writes the implementation, tests and docs; I set the brief, decide what ships, and validate it myself, including with a controller in hand for the game. Every status word above is exact, never rounded up, and a page changes the moment the status does. That's also why the game's documentation outweighs its code: most of it is context an agent reads at the start of a session, not marketing.
Where this would not fit
If you need hands-on-keyboard implementation from me personally, line by line, that's not how I build anymore — I direct agents rather than write the implementation. I write in Markdown, and if that counts as source code, I'm writing constantly, but I haven't written a line of TypeScript, Python or C++ myself (aside from the occasional CSV, JSON or YAML edit).
More on the work: /builds/. Every number: /receipts/.