# The autonomous dev pipeline: a product director's build system

**Status: in use on my team. Role: builder and operator.**

## What it is

An AI pipeline that develops software from our ticket queue with no human in the loop
until code review. It pulls a ticket, classifies it (backend, frontend, or web portal),
routes it to the right codebase, implements the change against our architecture standards,
writes tests, reviews its own diff, opens a labeled pull request, and updates the ticket.
Human engineers review and merge; the pipeline does the rest.

I built it. I'm a product director, not an engineer.

## Why a PM built this

Two reasons, one practical and one strategic.

Practical: my teams are small and the backlog of genuinely valuable small work (papercuts,
polish, config changes, minor features customers ask for) is endless. Every one of those
tickets used to compete with big bets for the same engineer-hours and usually lost.

Strategic: every product and engineering team needs to stay ahead of the frontier of what
AI can realistically build. Today that frontier covers smaller, well-scoped work. It will
keep expanding into larger and more complex projects. For years PMs had to sacrifice
incremental improvements to fund the big innovative bets. AI deletes that tradeoff: the
incremental layer increasingly takes care of itself, so teams pursue bigger bets without
letting the product rot. The best teams of tomorrow are already watching this frontier
today. Building the pipeline myself was how I made sure mine is one of them.

## What it changes about the job

The interesting shift isn't speed. It's that ticket-writing became the quality bottleneck:
a well-specified ticket comes back as a mergeable PR, a vague one comes back as a lesson
in specification. Which turns out to be excellent training for product managers.

This website, its fit-check agent, and that pipeline were all built the same way.

*Back to [ajsoltero.com](/) · [all case studies](/llms.txt)*