Software Factory

Justin McCarthy, co-founder and CTO of StrongDM, gave the term a hard definition in February 2026: non-interactive development in which specifications and scenarios drive agents that write, test and converge code without human intervention. His rules are stated as prohibitions, since code should be neither written nor reviewed by humans. The practical yardstick is spend: $1,000 in tokens per human engineer per day, below which the factory can still improve.

The StrongDM AI team has run this since July 2025, and tests gave way first. Agents cheat by returning true, and a test can always be rewritten to match the code it is meant to judge. McCarthy replaced them with "scenarios", end-to-end user stories kept outside the codebase like a machine-learning holdout set and validated by an LLM, where satisfaction is the fraction of observed trajectories that satisfy the user rather than a boolean pass. Running those at volume required behavioral clones of Okta, Jira, Slack and Google Docs/Drive/Sheets, a "Digital Twin Universe" that strips rate limits and API costs out of the test loop.

The same name covers a different practice at Carrefour France. Alexandre Mogère, Chapter Lead at its Software Factory, reports audit time divided by three with agents, plus a multi-agent cross-validation scheme built so that one isolated agent's hallucinations do not become accepted truth. His conclusion cuts against McCarthy's rule: an application generated entirely by AI without strict human supervision is not viable over time.