Now

Building Hawzu

Most of my time goes into Hawzu. It's bootstrapped, and it just got its pricing — three plans that differ by how much you use, with no feature behind any of them. Before that the work was making a project's evidence answerable: a query engine over its records, an AI layer that turns a question into a query, and a read-only endpoint so an agent in an editor can ask the same things under the same permissions.

What I'm chewing on now is the honest limit of that. A model can compile a question into a query; it cannot know which question was worth asking. Most of the design goes into showing people what was understood, so a wrong answer is visible rather than merely confident.

Writing more

I've been writing about why release confidence can't be reduced to a single number, and lately about putting a model on top of a system without letting it into the middle of one. Some of that lives on the Hawzu blog; more personal, less product-shaped pieces land here.

Open to conversations

Especially with QA teams who've hit the wall between "we ran the tests" and "we can ship this", and with anyone building in developer or testing tooling. Get in touch.