An AI lab for product managers who don't code

Follow a fixed learning path. Run probes against real models yourself, write down your judgment, and connect the abstract concepts from courses to the examples you ran into.

Probe workbench

Write your expectation first — only then can you run. Call models with your own API key, compare multiple runs side by side, and record the failure, its root cause, and how you'd work around it in a product.

Concept index

One card per concept: a one-line definition, a link to the original lesson, and your own words. Each concept page gathers every probe and judgment you tagged with it — that's your review page.

Judgment log

A few lines a week: the concept you learned, how your judgment changed, what's next. Append only, never rewrite — the change itself is the point.

Invite only for now. Built for PMs and product leads willing to spend 3–5 hours a week running experiments by hand.