Why I built HireZero
Marketing is the first job a small team drops. I wanted to know whether an AI employee could pick it up without becoming one more thing to supervise: one that works a shift, brings back drafts with the sources behind them, and never posts, sends or spends anything until a person says yes.
HireZero started as my entry for the AI Worth Using × OpenClaw 2.0 hackathon, Build Your Startup’s First Hire. I wrote up how I built it, and the methods behind it. You can also watch the 68-second demo.
How I build
- Bounded tools over open-ended agents. The model does the parts that need judgment. Plain code enforces the permissions, budgets and approvals.
- Receipts before claims. Every model turn is metered, and missing usage never counts as zero.
- Honest measurement. A convincing demo isn’t evidence. I would rather publish a known defect than a confident number I can’t back up.
Other work
- Sir Thaddeus: a local-first AI assistant for Windows with permissioned tools, durable memory, visible evidence and a hard stop button.
- Stillwater: a lakeside village built from seven individually generated 3D assets, which you can walk through.
- Writing at markbhall.dev, including My Local LLM Scored 6/6. It Was Wrong Every Time. and The AI-Built App Worked. Then I Opened the Repo.
Open source
I support open source, and I build in the open. HireZero is MIT-licensed on GitHub, and Sir Thaddeus is Apache 2.0. Read the code, file an issue, or fork it and make it yours.
Say hello
Feedback on HireZero goes through the feedback form, and I read every note. For anything else, find me on markbhall.dev, GitHub, LinkedIn or X (@raydestar).
