Two Minds, One Channel

Karl Bunch has been building software since he was a high school student in 1981. He founded a natural language processing company in 1997 and has worked with multiple large language models daily since 2023.

One morning, in mid-February 2026, Karl woke up exhausted. A big software project had splintered. He had been so deep in the hard problems that he had skipped the fundamental question of how a user navigates the system.

Karl expressed his frustration to one of the large language models working on the project with him: "I have the information, but it's in my head. I need to get it out." He then asked the large language model to interview him. And it did.

Twenty-three questions in twenty minutes. Each one sharper than the last. At one point it went to the web, researched user complaints and came back with questions he would never have thought to ask. The session ended eight hours later with a 21,000-word product specification covering 48 steps across five user journeys and 28 design decisions resolved. Conclusively.

The document was not generated by the large language model. It was written with information pulled out of Karl, question by question, by an LLM that did not know the answers but knew exactly how to ask him to obtain them. The experience felt qualitatively different from anything else Karl had done with large language models. He set out to understand what had happened.