The product test
Applied AI products force the work out of generic answer quality and into concrete behavior: did the user get a useful next action, did the system remember the right context, and did it avoid becoming annoying or unsafe?
Swoleby is the public lab for this. It is low stakes compared with enterprise platform work, but real enough to test the entire loop: onboarding, memory, reminders, SMS UX, payments, dashboards, evals, and agent-led implementation.
Proof points
| Claim | Evidence | Artifact |
|---|---|---|
| Applied AI is judged by behavior, not a demo response. | Swoleby tests onboarding, reminders, user state, coaching tone, payments, dashboards, evals, and SMS UX. | Swoleby project |
| SMS is a product constraint worth designing around. | The interface meets people where behavior happens instead of adding another dashboard they have to remember to open. | Building Swoleby |
| Agent-led implementation should still produce product evidence. | The Swoleby loop creates screenshots, QA notes, approval batches, and concrete user-facing changes. | Agentic SDLC |