01
Scope with AI, decide with people
We use AI to explore the problem space fast — competitor teardowns, data-model options, edge cases nobody listed. The architecture decision is made by an engineer and written down.
Gate: Architect signs the design doc
02
Generate the skeleton
Project setup, data models, CRUD, API contracts, and typed clients are generated against our own templates rather than typed out for the hundredth time.
Gate: Lead engineer reviews the scaffold
03
Build the parts that matter by hand
Business logic, pricing rules, permissions, anything with money or safety attached — written and reasoned about by a senior engineer.
Gate: Peer review, no self-merges
04
Draft tests with AI, harden them with QA
AI proposes the happy paths and obvious edges; QA adds the cases only someone who has watched real users would think of.
Gate: Coverage delta must not fall
05
Review every diff, generated or not
One checklist for all code: tests, dependency diff, licence scan, threat model impact, and a named human approver on the merge.
Gate: Named reviewer recorded per merge
06
Ship, watch, and iterate
CI/CD, observability, and a weekly demo. Post-launch, AI helps triage logs and draft fixes — the fix still ships through the same gate.
Gate: Release sign-off by delivery lead