What AI Agents Change About the PM Craft, and What They Do Not

There is a Lenny's Newsletter episode with a title I love, make product management fun again with AI agents, and the core idea has stuck with me: agents handle the execution, humans handle the judgment. It comes with a practical playbook, spot a low-risk, high-impact task to automate first, and a five-question checklist to keep the agent safe, scoped, and effective, controlling cost and curbing hallucinations along the way. It frames the agent as an intern that earns its way to trusted teammate. That framing is right, and it maps cleanly onto healthcare product work.

Here is my version of it. AI agents change what a product manager does with their hands, not what they do with their head. The execution, the research pulls, the first drafts, the data cuts, moves to the agent, and it is genuinely faster there. The judgment, what to build, what matters, what to say no to, stays human, because that is the actual job. In healthcare I hold that line even tighter, because the decisions that matter are regulated and clinical, and an agent cannot own them.

Execution moves, judgment stays

The clean way to think about agents in product work is to split every task into execution and judgment, and only delegate the first. Almost everything a PM does has an execution half and a judgment half. Research has the pulling and summarizing, and the deciding what matters. A spec has the drafting, and the point of view. An analysis has the running, and the read. Agents are excellent at the first half of each and cannot do the second. So I delegate the execution and guard the judgment, and I never let the speed of the execution half fool me into thinking the judgment half is done. The table below is where I draw the line, and the clinical row is the one I never move.

Delegate without losing control

Delegating to an agent safely is a skill, and it looks a lot like delegating to a sharp new hire you do not yet fully trust. The mistake is to hand an agent the important judgment call because it is fast and confident. The discipline, close to the checklist Lenny describes, is to start small, scope tightly, guard against cost and hallucination, keep a human on the decision, and widen only as trust is earned. In healthcare I add one rule that never bends: anything regulated or clinical stays with a person, no matter how good the agent looks. Speed on the execution is a gift. Speed on the judgment is a trap.

I am genuinely more optimistic about the PM craft because of agents, not less. They take the grind, the pulls, the drafts, the cuts, off my plate and hand it back done, which leaves more room for the part that was always the point: discovery, prioritization, judgment under uncertainty, the four risks that decide whether a product is worth building. Let the agent make you faster at execution. Stay the one who decides. And in healthcare, keep the regulated and clinical calls firmly in human hands, where they belong.