Using Coding Agents to Build Reliable AI Agents

Building an AI agent that works in production is a systems problem, not a prompting problem. In this talk at the O'Reilly Software Development Superstream "From AI Conductor to AI Orchestrator" I show a workflow built on two agents: a coding agent, which writes and updates the code base, and a product agent, which is the tool you ship to the end user.

We look at how the two work together, how self-healing evals let the coding agent find and fix the product agent's mistakes, and how to keep the whole loop reliable as it grows.