A recent thread in Lenny's Slack community asked experienced product managers whether AI had made them feel more replaceable or less. Everyone commenting was experienced, and the answers were nearly unanimous: less replaceable, more capable. AI had taken the commodity work, and judgment had gotten more valuable.
Every senior PM in that thread had built their judgment through years of friction-filled work that AI now does in seconds, and the assumption underneath the conversation was that newer practitioners would somehow find their way to the same place. I don't think that happens by accident anymore, and believe the best way to train new PMs is to add back some of that friction.
What the friction actually did
The manual work wasn't just inefficient, it was its own form of training. Sitting with hundreds of support tickets taught a PM how customers think. Puzzling through a competitor's product decisions taught a PM how to read a market. Writing a spec from scratch forced a PM to understand what was actually being built well enough to explain it to someone else.
AI handles all of that now, faster and often better than any individual. The synthesis, research, first-draft specs that AI tools create are a huge time saver for organizations. But what those tools stripped out in the process is the struggle that built judgment in the first place, and the risk isn't that newer PMs have worse tools. It's that the tools are so capable they allow someone to skip the work that teaches them when to trust the output and when to push back on it.
That's the skill gap quietly accumulating inside product organizations right now, and most of them aren't designing for it.
How experienced organizations can reintroduce the friction
The approach I've been advising is to bring a newer PM into the human-in-the-loop eval process, not as an observer, but as an active participant working alongside a more experienced practitioner.
The senior PM should establish evaluation criteria for the agent's output: what good looks like, and what would prompt them to push back and update the prompt. The junior PM is there to learn the judgment, not just the mechanics. That means asking why, not just watching what.
Over time, the newer PM takes on more of the evaluation work directly. They start making calls on the output and defending those calls with specific data, not just an instinct that something seems off. If they approve the output, they have to articulate what they're watching for in future runs that would signal the loop is drifting. With any rejections, they have to explain what needs to change in the prompt and why. The experienced PM's role shifts from doing to probing, asking questions that require the newer practitioner to demonstrate their reasoning rather than just their conclusion.
What forces skill growth is the requirement to close the loop. The newer PM makes a judgment, forms a hypothesis about what change will improve the output, makes the change, and then documents whether quality actually moved. That cycle, judgment to hypothesis to measurable result, is the work that builds strategic thinking. It's also the work that AI tools cannot do for someone.
The goal is for the junior PM to eventually own the eval entirely, with enough context and judgment to run it, defend it, and update it without needing the more experienced practitioner in the room. That transition doesn't happen on a fixed schedule. It happens when the newer PM has demonstrated, repeatedly and with evidence, that their judgment can be trusted at each stage before moving to the next.
The harder part
This only works if an organization sees the value in training product managers, and training senior product managers how to design human-in-the-loop evals that a newer PM can eventually own. The old apprenticeship model was slow and inefficient, but it paid people to develop judgment over years. The new one requires established PMs to deliberately reintroduce friction that the tools just removed, and that doesn't happen by default. It requires a choice to slow down slightly, create a structure where a newer PM has to do hard cognitive work, and hold them accountable to it over time.
Most organizations aren't doing this. The incentive is to capture the speed AI offers and move on. But if every experienced PM optimizes only for their own output, the institution runs out of product managers with the skills an org needs in a very short time. The path to developing strong product practitioners changed. The requirement for a company to build those skills within their teams did not.