StrongDM's AI Team Builds Software Without Human Code

StrongDM's AI team has achieved a breakthrough in software engineering by building functional systems without any human-written code. Formed in July 2025, this three-person teamJustin McCarthy, Jay Taylor, and Navan Chauhanadopted a radical rule: no hand-coded software. By October 2025, they demonstrated working prototypes of coding agent harnesses and clones of major services like Slack and Okta.

This approach aligns with what industry observer Dan Shapiro terms the 'Dark Factory' level of AI adoption, where humans oversee but do not touch the code directly. The team's success leverages advancements in AI models like Claude Sonnet 3.5, with further boosts from later releases such as Claude Opus 4.5 and GPT 5.2 in late 2025, which improved agent reliability for complex tasks.

Central to their method is the 'Digital Twin Universe,' independent clones of production services like Slack and Okta. These clones operate without rate limits or quotas, enabling swarms of simulated test agents to run scenarios continuously. For instance, a Slack twin simulates streams of Okta users requesting access, allowing rigorous testing as systems evolve.

Creating high-fidelity replicas of SaaS apps was technically possible before but economically unviable. AI agents now make it feasible, disrupting traditional development by providing unlimited testing environments.

The team introduced specialized techniques to enhance agent efficiency:

• Gene Transfusion: Agents extract patterns from existing systems and apply them elsewhere.
• Semports: Direct code porting between languages.
• Pyramid Summaries: Multi-level summaries for quick agent navigation, from high-level overviews to detailed info.

These tools allow agents to handle complexity scalably.

Monty enables safe execution of LLM-generated Python code with microsecond startup times, bypassing slow container sandboxes. This reduces latency and cost, making agent-driven coding practical for real-time iteration.

StrongDM specializes in secure access management, providing just-in-time (JIT) runtime authorization for humans and non-human identities (NHIs). Their platform supports cloud-native, hybrid, and on-prem environments, enforcing least privilege at the point of action.

In January 2026, Delinea announced its acquisition of StrongDM to combine privileged access management (PAM) with JIT authorization. The merged platform targets the agentic AI era, offering real-time control over AI agents, reduced credential exposure, and frictionless developer access to resources like databases and CI/CD pipelines.

This positions StrongDM's innovations amid growing demands for securing AI-driven workflows, where NHIs outnumber humans and require zero standing privilege (ZSP) models.

The AI team's demos preview a shift where small teams oversee AI swarms building enterprise software. This could accelerate development cycles, lower costs, and enable bold testing previously shelved due to resource constraints.

As AI models like GPT-5.2-Codex see widespread adoptiondoubling usage post-launchtools like those from StrongDM's team will integrate into broader ecosystems. Their work complements StrongDM's Leash project, an open-source tool for real-time policy enforcement in agentic AI, enhancing visibility and safety.

Challenges remain, such as ensuring agent reliability and integrating with legacy systems, but StrongDM's approach demonstrates viable paths forward. For developers, this heralds an era of supervisory roles over autonomous coding factories.