AI-augmented engineering lets a single operator carry the velocity of a team. But leverage is only as good as the judgment directing it.
One operator orchestrates parallel specialist agents — the velocity of a team, held by a single pair of hands.
Code review and evals stand between generation and production. Nothing ships on vibes.
The decision point is a person — me. I answer for what ships, not the agent that wrote it.
Manual test first — if I haven't seen the code do the right thing, it doesn't work. Then automated tests that fail if the change is reverted.
RLS, auth, access rules — never outsourced to the agent. Knowing what counts as security-adjacent is the skill.
The depth to read the code and reproduce a fault beats re-prompting until it looks fixed.
For anything AI-shaped, a small eval suite catches the failure modes before users do. Read the write-ups →
A reusable tool extracted from the private products as a generic by-product — open the framework, keep the product. Candidate: an RLS-policy testing harness or a migration-linting utility distilled from ~150 Supabase migrations.