Person/Simon Willison/Blog/26/02/How StrongDM's AI team build serious software without even looking at the code

• How StrongDM's AI team build serious software without even looking at the code

  • My Notes

    • StrongDM describes a "software factory" workflow where humans focus on specs, scenarios, harnesses, and system constraints instead of manually writing implementation code.
    • The sharpest claim is the AI/Coding/Dark Factory principle: code is neither written nor reviewed by humans, and quality is enforced through scenario coverage and empirical "satisfaction" metrics.
    • Their StrongDM/Digital Twin Universe idea is especially interesting: build high-fidelity local clones of third-party systems (like Okta, JIRA, and Slack) to safely run thousands of scenario checks per hour.
    • The release strategy for StrongDM/GitHub/attractor is a notable spec-first pattern: publish the complete markdown spec and let coding agents generate the implementation.
    • StrongDM/GitHub/cxdb looks like a practical "context substrate" for agentic work, with immutable DAG-based storage for tool outputs and conversation history.

  • Highlights

    • "Code must not be written by humans"
    • "Code must not be reviewed by humans"
    • "If you haven’t spent at least $1,000 on tokens today per human engineer, your software factory has room for improvement"
    • "We repurposed the word scenario to represent an end-to-end ‘user story’, often stored outside the codebase (similar to a ‘holdout’ set in model training)."
    • "With the DTU, we can validate at volumes and rates far exceeding production limits... run thousands of scenarios per hour without hitting rate limits."