Software Factory

Software factory describes a system of work where coding-agent sessions are started by triggers rather than by a person: an issue being created or labelled, a cron schedule, a CI failure or monitoring alert, or another session finishing. The term appears in Matt Pocock's AI Coding Dictionary (aicodingdictionary.com, open-sourced on GitHub as mattpocock/dictionary-of-ai-coding), a glossary of agentic-coding jargon. Without a factory, every agent session still needs a person to open it and point it at a task, even when the work that follows runs AFK (away from keyboard, unattended). A factory removes the human from that one step of starting the session. It does not necessarily remove the human from any step after it.

A factory does not have to cover an entire development process to count as one. The dictionary's own example is a single cron job that fixes one lint-rule violation a night and opens one reviewable pull request. A narrow loop is enough. Reviewing its output shows how far it can be trusted before it is widened to more triggers or more kinds of sessions. Other common triggers include a failing build on the main branch starting a session that finds the breaking commit and proposes a fix, or one agent's pull request triggering an automated review whose comments start a fix-up session.

Humans can sit anywhere in a factory: writing and labelling the issues that trigger sessions, approving a plan before implementation starts, or reviewing a diff before merge. Deciding which of those checkpoints stay human, and which do not, is the main design question a factory raises. The dictionary names the limiting case, a codebase or part of one where no human ever reviews the factory's output, "dark factory."