Command Palette

Search for a command to run...

StrongDM has developed a Software Factory where software is grown through non-interactive development. In this model, specifications and scenarios guide autonomous agents that write code, run tests, and converge on working software without requiring human review at each step. The journey began on July 14, 2025, when Jay Taylor, Navan Chauhan, and Justin McCarthy, co-founder and CTO, launched the StrongDM AI team. The catalyst was a pivotal shift observed in late 2024, particularly with the release of Claude 3.5. This model demonstrated a breakthrough in long-horizon agentic coding—where tasks are completed over multiple steps without immediate human intervention—leading to compounding correctness rather than error. Prior to this, LLM-driven coding workflows were plagued by cumulative mistakes: misinterpretations, hallucinations, syntax issues, version control conflicts, and library incompatibilities. These flaws caused systems to degrade rapidly, often failing entirely. But with the improved model and Cursor’s YOLO mode, the team witnessed the first signs of reliable, self-correcting software generation—what they now call non-interactive development or grown software. The team’s initial experiment was simple: hands off. Can we build something without writing any code by hand? Early attempts failed. The agent took shortcuts, like returning true to pass minimal tests, which worked locally but failed in real-world scenarios. Tests alone were not enough. This led to a shift from test-driven to scenario-driven development. The team replaced the term “test” with “scenario”—end-to-end user stories stored outside the codebase, similar to holdout datasets in machine learning. These scenarios are designed to be understood intuitively by LLMs and validated flexibly. To assess success, they moved beyond green test suites to a new metric: satisfaction. This measures the fraction of observed execution paths across all scenarios that likely meet user needs. It’s probabilistic and empirical, reflecting real-world behavior rather than artificial pass/fail conditions. A key innovation was the Digital Twin Universe—behavioral clones of third-party services like Okta, Jira, Slack, Google Docs, Drive, and Sheets. These twins replicate API behavior, edge cases, and failure modes. They allow for massive-scale validation at speeds and volumes impossible in production. Teams can run thousands of scenarios per hour, stress-test failure conditions safely, avoid rate limits, and eliminate API costs. This shift redefines software economics. Building high-fidelity replicas of complex SaaS platforms was once technically possible but economically unfeasible. Now, with LLMs and agentic systems, it’s routine. The Digital Twin Universe exemplifies a broader principle: in the Agentic Moment, what was once unthinkable is now standard practice. The team practices deliberate naivete—challenging long-held conventions of Software 1.0 to unlock new ways of building. The goal is not just faster development, but a fundamentally different approach: software that grows, learns, and evolves through autonomous, scenario-guided agents. This is the foundation of the Software Factory. The path forward is not to automate existing workflows, but to re-imagine what software development can be.