Agentic Engineering Playbook

Case Study: Billi (in progress)

Billi is the first real project we will scaffold with prototype-kit. The four-week MVP cycle begins 2026-04-20; this page will update with actual outcomes as the cycle runs.

Case Study: Billi

Billi is a personal finance app designed for the Mexican market — specifically for university students aged 18-25 who want to track daily expenses without the complexity of traditional budgeting tools.

This case study is in progress. The Billi MVP build cycle begins 2026-04-20 and runs for four weekly cycles ending 2026-05-17. Below we state the hypothesis we are testing; we will update this page with observed outcomes after each cycle.

Hypothesis we are testing

A student with no development background can go from three short markdown docs (product-brief.md, ux-vibes.md, screens.md) to a running React prototype in the same afternoon by using the prototype-kit workflow described in this Playbook. Specifically, we expect:

  • Documentation-to-first-render in under three hours.
  • A prototype polished enough to demo in front of a jury without apology.
  • Less than one manual intervention per generated screen.

What we expect to learn (in priority order)

Where product-brief.md specificity actually matters. We will track which brief sections shape the generated UI most: the problem statement, the target user, the MVP feature list — and compare outputs when any one of them is vague vs. specific.

Whether ux-vibes.md outperforms hex codes. We will deliberately run one iteration with only reference sites and tone adjectives, and one with hex codes and typography specs, then compare which produces a more coherent visual result.

How the agent handles ambiguity. We will note every time the agent asks a clarifying question versus fills in a plausible default — and whether the defaults were acceptable.

Status

Week 0 (planning) is complete. Cycle 1 begins 2026-04-20. This page is scheduled to update at the end of each weekly cycle.