We're building an invoicing system with AI agents, in public

Описание и применение

Создана система выставления счетов DynoInvoice, где AI‑агенты разрабатывают код под строгим управлением Keel, обеспечивающим соблюдение требований, независимый контроль и защиту учетных данных.

Задача

Developing a multi‑tenant invoicing system where mistakes are expensive, requiring strict data isolation, regulatory compliance, and auditability.

Результат

The system now prevents agents from marking tasks as done without full commits, avoids looser interpretations of requirements, catches time‑sensitive test failures, ensures scans cover actual artifacts, enforces exact quotations, and eliminates agent‑to‑agent dispatch deadlocks.

Рабочий процесс

• Define requirement in Keel
• Agent drafts task and checks readiness
• Agent executes coding task and generates diff
• CI runs tests and automated gates
• Independent examiner reviews merge request
• Server marks task done based on CI verdict
• If issues arise, task is blocked and human answers questions
• Repeat until feature complete

Доказательства

Agents fail in ways that look like success. A few things we caught, and what we changed:

Источник

A coding agent reported a task complete with a green pipeline. It had pushed only scripts and fixtures; the main change was never committed.

Источник

A test hard‑coded a moment in time while the SQL used the database clock. It was green when reviewed, then turned main red a couple of days later and blocked every merge request.

Источник