The PRD was built for software that does what it's told. An agent exercises judgment, and the moment it crosses from suggestion to decision, the old brief breaks. It splits in two: one document to decide, one to build. The new AI tools can write the second. The first is still yours.
Second edition. The judgment a person used to supply at runtime, and how to write it down before the agent runs. Fifteen chapters for PMs shipping agentic AI where being wrong has consequences.
I had undiagnosed ADHD for most of my life. Lecture halls were the wrong delivery mechanism for me. So I went to the library, ran my own research, built my own curriculum. Six months ago, the constraint of scale disappeared. The library is still the right model. The library is just bigger now.
The senior engineer approved nineteen pull requests today. She read four of them. The supervision channel is collapsing in real time, and the validator is atrophying as the validated work grows. We have fifty years of HR experience supervising smarter reports. None of it has reached the agent.
Every model upgrade is a personnel change nobody approved. The persona prompt is the costume. The model is the actor. The actor wins when the stakes get high or the prompt gets thin. We have fifty years of personality assessment for human hires. We have not pointed any of it at the agents.
We run every new hire through recruiting, references, a probationary period, and a 90-day review. Then we deploy an AI agent into the same workflow and skip every gate. The technology question is settled. The HR question has not been asked.
Medicine spent twenty years building governance frameworks for finding-first medicine. Fleischner. Bethesda. Lung-RADS. Each took years, named authors, bounded scope. Consumer health AI synthesizes your entire medical record before you see your doctor. Nobody built the framework for that.
Clinical reasoning was built on a simple sequence: observe, hypothesize, test. Three disruptions have quietly inverted it. The signs still work. The sequence that gave them their meaning is disappearing.
The judgment layer just shipped. A patient arrives with a Google Health Coach synthesis of her last two years of medical records. Her physician receives a pre-processed narrative - no author, no methodology, no version history. The default answer is now Google.
Your dashboards are green. Your agent approved 17 wrong purchase orders overnight. Traditional O&M answers "is it running?" Agentic O&M must answer "is it behaving correctly?" These are different questions. They require different instruments.
Your agent passed every security check. The tools your team used were built for a different system. The frameworks that cover agentic AI are months old, the enterprise adoption cycle is 12 to 18 months long, and the models getting better at finding your gaps ship faster than your procurement cycle.
LLMs are getting faster, cheaper, and better every week. The comparison charts prove it. They also prove nothing about who wins. I've watched six technology wars end. The best product lost every time. The winner owned the right layer of the stack.