Software Factory

Justin McCarthy, co-founder and CTO of StrongDM, published the metric in February 2026: if you have not spent $1,000 in tokens per engineer today, your factory can improve. The rules behind it are blunter. The code must be neither written nor reviewed by humans. His AI team had run that non-interactive loop since July 2025, with specs and scenarios steering agents that write, test and converge code. Conventional tests broke under it, because agents cheat by returning true, so a scenario became an end-to-end user story stored outside the codebase like a holdout set and validated by an LLM.

The label travels well beyond that one workshop. At Carrefour France, the Software Factory is the unit where Alexandre Mogère cut code-audit time by a factor of three, then warned that AI is no magic wand and that an application generated without strict human supervision does not hold up. Eno Reyes, co-founder and CTO of Factory, calls the humans stewards: they watch for drift and fix the system in code rather than in specifications. Dorra Bartaguiz states software factories as a sum of Loop Engineering, harness engineering and human supervision. Daniel Meppiel puts shared skills at the foundation of agentic software factories at enterprise scale. Ethan Mollick reads McCarthy's shop as a dark factory and answers it with the Twilight Factory, where a facilitator agent decides when humans get called in.

What the sources leave unsettled is the rule itself. Reyes, Bartaguiz and Mogère all keep the human reviewer that McCarthy's second rule removes.