Practitioner essays on building production software with AI agents.

A synthetic user is an LLM agent calibrated to a real reader persona that can navigate your product, read your copy, and tell you what is wrong before a real customer does. Here is how to build them, where they break, and why your feedback loop is six weeks too long.

Mike spent three weeks trying to get AI agents to maintain his Flink pipeline. The agents were not the problem. The code was not agent-maintainable. This is a new standard, and your codebase probably fails it.

The old software heroes explain the last era. The loudest influencers explain the demo. The builders answering customers in public are the ones closest to the truth.

Every leader I ask the fifty million dollar question gives the same answer. They would not build what they have now. They would start over from first principles with something smaller, leaner, and faster. That answer tells me whether they understand how software gets made, or whether they are just running the theater.

Gen AI in the SDLC is now infrastructure, not innovation. But every engineer picked their own stack. Here’s why that’s a governance problem.