What It Actually Takes to Build a Software Factory — Tereza Tížková, Factory
Source
AI Engineer
Author
AI Engineer
Date
Key takeaways · AI-distilled
Tížková defines a software factory as the whole lifecycle run autonomously, from collecting signals and prioritizing through building, validating and improving, and says writing code is the easy part.
Factory's worker agents run in sequence rather than as a swarm, so each starts with fresh context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition →, and separate validators check code they did not write, one of them by clicking through the running app.
She cites about 25% savings from automatic model routingSending each request to a model chosen by the difficulty of the task, rather than using one model for everything.Full definition → on Factory's conservative benchmarkA standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.Full definition →, and says a deferred context engine cuts token use by 50% or more.
Factory's Missions run for hours or weeks; she cites a real customer run lasting 16 hours, and an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition →-readiness check aims to stop AI from making a messy codebase worse.
Terms in this piece · Glossary
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
model routing — Sending each request to a model chosen by the difficulty of the task, rather than using one model for everything.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Why it matters
Lays out specific techniques for running autonomous software factories at scale, including a context engine cutting token use over 50% and validators that actually exercise the running app, not just read the diff.
