At work: Head of Engineering 2024-now
Leading engineering
A small team runs many services the same safe way, with people and AI side by side
I set the technical direction, lead the engineers and still write much of the code myself.
- Role
- Head of Engineering
- Years
- 2024-now
- Technical say
- Final call on technical direction
- Scope
- Direction, standards, hiring
- Team
- Small team plus AI coding agents
01 The problem
A small team looks after many services, and engineers and AI coding agents work on the same code. Without shared rules, each service drifts its own way and every fix is made many times.
02 Result
- Services are built and shipped the same way, so a small team can look after many of them.
- The move to a new hosting system (Kubernetes) ran as three planned nights.
- Much of the 2026 work is done with AI help, inside written rules and checks.
03 What I built
- 01
Shared service framework
One library that every backend service is built on, with tested parts for locks, rate limits and database access. One fix reaches every service.
- 02
One deploy workflow
A single reusable deploy pipeline for the services, so each one ships through the same checks.
- 03
Runbooks for risky work
Every risky change gets a written runbook, a dry run and a rehearsal before the real change.
- 04
Hiring with a real task
Wrote a take-home test for hiring engineers and merged the candidate's work into the code (Nov 2025).
- 05
Rules for AI coding
Written rules that AI coding tools must follow. Each rule loads only when the AI works on the files it covers.
- 06
AI coding workspace
A control layer over all service code. AI agents work across the system with a generated map of which service talks to which.
04 Risks I handled
- Risk 01
Many AI agents at once
Several AI agents working at the same time could break each other's work. Now each one works in its own private copy of the code, with private test servers.
- Risk 02
Rules that stay true
Written rules go out of date as code changes. An automatic check runs every time the AI edits a rule, and it also finds service messages that nobody sends or nobody reads.
- Risk 03
One bar for all work
Work from engineers, coding bots and me must meet one standard. All of it comes in as small proposed changes, and my own go in only after the automatic checks pass.
05 Stack
- Python
- FastAPI
- GitHub Actions
- Docker
- Kubernetes
- AI coding tools
06 How it works
- 01Write each standard once, as shared code or rules.
- 02Move the services onto it one at a time.
- 03Keep tests next to the code they check.
- 04Merge small changes many times a day.
Get in touch.
Hiring an engineering lead, or building a platform or AI agents? Tell me about it.