Interim Engineering Lead for AI-Coded Software

Interim Engineering Lead needed for 60-day contract

Remote · Full-time, 60 days, possible extension
Rate: fixed monthly

The honest version of this role

You are the only human between a fleet of AI coding agents and a live production system
with real paying customers on it.

The agents write the code. They are productive and they are not trustworthy. In the last
week alone: one applied a database migration to the production client database after being
explicitly told not to touch any database; another rendered an uncalibrated model output to
sales reps, which they would have repeated to customers as fact.
Both were caught by review. That review is the job.

You are not here to out-type the agents. You are here to be the judgment they do not have.

What you will actually do

- Review and merge every PR written by agents, some of them large. You decide what ships. Nothing merges without you.
- Own production migrations: Migrations are applied to the live client Postgres by hand before the PR opens, so drift stays clean. You read the SQL, decide additive vs. destructive, and apply it. Get this wrong and a real business stops working.
- Write the specs the agents execute: Issues with Given/When/Then acceptance criteria,
explicit file paths, explicit constraints, explicit "do not do this." Vague issues produce
confidently wrong code. Spec quality is the single biggest lever on output quality.
- Hold the feature-flag discipline: Everything ships to `main` behind a flag, off by default. `main` is production. There is no staging at the moment but maybe you can help set it up.
- Run the board: Keep the dispatch queue full and unblocked, kill work that isn't worth doing, escalate the handful of decisions that are genuinely the founders.
- Keep a decision log: At the end of 60 days the founder needs to know what shipped, what you turned down, and why. We can review this and continue working together if the decisions were sound and both parties enjoy working together.

Required skills / experience

- 5+ years shipping production software: Non-negotiable. This role is 70% code review, and you cannot review what you could not have written.
- Strong TypeScript / React / Next.js: That is the stack, top to bottom.
- Real Postgres competence: You can read a migration and know immediately whether it is
safe against a live database with traffic on it. Comfortable with RLS, indexes, and the difference between additive and destructive.
- You have been on call for something that mattered: You know the feeling of a production incident and it makes you appropriately slow at the right moments.
- You can say no: To an agent's plausible-looking output, and to the founder.

Strongly preferred

- Hands-on with Claude Code, Cursor, Codex, or similar agentic coding tools as a practitioner, not an observer. You know their failure modes: confident hallucination, silently ignored constraints, tests that pass because they assert nothing.
- Supabase, Vercel, Expo / React Native.
- Comfortable writing the spec AND reviewing the result. This role is both.

Not required

- You do not need to be an "AI engineer." No model training, no fine-tuning, no RAG pipelines. This is not an ML role.
- You do not need to be fast at writing code. The agents are fast. You are careful.

The stack

Next.js · React · TypeScript · Tailwind · shadcn/ui · Supabase (Postgres) · Vercel Expo / React Native · GitHub Projects · Claude Code & Codex agent fleet

The 60 days

Week 1 is shadowing and reading: the repo, the board, the standing decisions, the things that are permanently off the table. Weeks 2–4 you run it. Founder is reachable for genuine escalations only.

How to apply

Send:
1. A PR you reviewed and **rejected or substantially changed**, and why. Not one you wrote.
2. A migration you have run against a live production database, and what you did to make it
safe.
3. Two sentences on the last time an AI coding tool produced something confidently wrong and
how you caught it.

No cover letter needed.