Restoring Trust in AI-Native Development

Autonomous Agent Loop with Poka-Yoke, Andon and Jidoka
AI coding agents have fundamentally changed the economics of software development. We can now generate code, APIs and entire applications faster than ever before. But while the forward loop of software delivery has accelerated dramatically, our verification systems haven’t kept pace.

This isn’t a new problem. Decades ago, manufacturing faced the same challenge. As production speed increased, adding more inspectors only created bottlenecks. Toyota solved this with three revolutionary principles: Poka-Yoke (mistake proofing), Andon (stop the line), and Jidoka (automation with human intelligence). These ideas quietly shaped modern software engineering from compilers and CI/CD to type systems and automated testing.

In this keynote, we’ll connect Lean Manufacturing, distributed systems, APIs and AI agents into a single narrative. You’ll see why correctness escaped the compiler boundary when we adopted microservices, why traditional code reviews and testing become the new toll booths in AI-native development, and how executable contracts restore compiler-like confidence to distributed systems.

If AI is the new power loom for software engineering, this talk explores what it takes to build the equivalent of Toyota’s self-stopping loom, where autonomous generation is matched with autonomous verification, preserving trust at machine speed.

Building on ideas from Harness Engineering, Naresh introduces four new capabilities: Executable Intent, Executable Architecture, Executable Specifications and Continuous Governance, which together with Harness Engineering, form Closed-Loop Engineering: enabling AI agents not only to generate software autonomously, but also to continuously verify it from intent through to the final architecture running in production.