I asked two senior leaders the same question within 24 hours last week, a CPO at a B2B software company and a CEO speaking to a room of product leaders at an industry event. Neither had an answer. In fact, one of them told me she'd "love 24 hours to think about it." The other said, bluntly, "I've never cracked it."
The question: how do people develop mastery when AI is doing the work that used to build it?
The CPO was walking me through a business case that had landed on her desk a few weeks earlier. Someone outside her product organization, well-intentioned and eager to demonstrate their value, had spent 40 hours with three different LLMs and produced a slide deck that looked like a startup pitch, complete with clean branding, a polished narrative, and detailed market sizing. But as soon as she started reading it, she could see it was hollow. The $3 million platform estimate was fantasy. There was no admin functionality, no entitlements, no account management, no back office. "It's not that AI hallucinated on this person," she told me. "It's that they didn't know what to ask it."
"It's not that AI hallucinated on this person. It's that they didn't know what to ask it."
Her diagnosis was sharp. "You still have to have mastery of knowledge," she said. "Particularly on the product side." She wasn't worried about AI making things up. She was worried about the people using it no longer having the judgment to know when something was wrong, or what to push back on, or what question to ask next. When I asked her how she thought about developing that judgment in people earlier in their careers, she paused. "How does somebody who's 25 today, or 30, or even 35... how are they going to have mastery of knowledge at some point? I've never cracked the how do you coach that, train that, or hire for it."
The next day I was at an industry event, sitting in a session led by a CEO in the developer tools space who was talking about how the manager's role is changing in the AI era. She walked the room through the data on what some are calling the Great Flattening. Manager layoffs tripled over the past three years. Middle manager hiring dropped 43 percent since 2022. Twenty percent of companies are expected to use AI as the justification to remove half their middle management by the end of 2026. Block cut a thousand employees in 2025 and explicitly eliminated 80 manager roles. Jensen Huang has 60 direct reports and doesn't do one-on-ones. Jack Dorsey reportedly wants all 7,000 employees at his company reporting into only three roles.
Her reframe of the manager's job was compelling. Focus shifts from tasks to impact. Role shifts from quarterback to coach. Responsibility shifts from overseer to system designer. Authority comes from judgment and trust rather than title and information. Her closing line landed hard: "Your job is not to manage the team's work. Your job is to manage energy, belief, and clarity of direction."
During the Q&A, I raised my hand and asked the question I had been sitting with since the call the day before. If the known answers are now table stakes, and the job becomes pushing for trade-offs, judgment, and taste, how do we ensure the people coming up behind us actually develop mastery? Those of us who have been doing this for 25 years built it through the school of hard knocks, through bad presentations and being grilled until it was clear I should come back when I knew more. The next generation will not get those reps the same way. How do we transmit what we learned? Her response was honest. "I'd love 24 hours to think about that question."
Two senior leaders, two industries, two days apart, and the same unanswered question in both rooms.
It is tempting to read this as a coaching problem. The narrative almost writes itself. Middle managers used to develop their people, now middle managers are being eliminated or stretched too thin to coach, therefore the next generation will not develop. But I don't think that is actually what is happening, and it is worth being honest about why. Middle management was never a reliable way to transfer mastery. Most middle managers I have worked with over the past 25 years were too busy putting out fires, managing up, and translating strategy into tasks to do meaningful coaching. When I think back on where my own judgment came from, very little of it came from a manager sitting me down and teaching me. It came from reading the 600-page MediaFLO spec cover to cover because I had no choice. It came from presentations that went badly. It came from business cases I defended in rooms where people who knew more than I did pointed out what I had missed. The school of hard knocks was not a backup plan for when coaching failed; it was the actual mechanism.
What is changing is not the transfer layer, it is the raw material. AI does not just accelerate the work. It removes the productive friction the work used to contain, and that friction was where judgment was built. The bad first draft that someone would have had to write, think through, and defend is now produced instantly, polished, and formatted. The 40-hour business case the CPO described is the perfect illustration. The person who built it did not struggle, they orchestrated. They never hit the moment where they had to think through whether account management and entitlements belonged in the platform estimate, because the LLM handed them something that looked finished before that moment could arrive. The struggle that would have built their judgment was compressed out of the work.
"AI does not just accelerate the work. It removes the productive friction the work used to contain, and that friction was where judgment was built."
Both leaders I spoke with see the problem clearly. Yet surprisingly, both of them are also, in the course of running their organizations, doing things that make it worse. The CPO is replacing engineers who are not AI-native with engineers who are. That is a reasonable operational decision, and I would probably do the same. But the engineers who leave take with them the experience of having built and debugged systems the hard way, and the engineers who arrive will be working in an environment where AI compresses much of the struggle that produced that experience. The CEO is advocating for fewer one-on-ones, more direct reports per manager, and pushing conversation into public channels. There are real arguments for each of those moves. But each of them also reduces the surface area for the kind of extended, messy, contextual exchange that transferring mastery actually requires. Neither leader is being hypocritical. The operational logic of running a modern organization simply has no obvious place to put the mastery problem, so it gets absorbed into the ambient cost of running the business.
This is not a new pattern. It is the acceleration addiction I have written about before, applied to people. Every previous wave of productivity technology promised to free humans up for higher-value work. Every previous wave ended with organizations taking the efficiency dividend and spending it on scale. AI is following the same script. The admin work that used to fill a manager's day is being automated, which in theory should create space to actually develop people. In practice, that space is being spent on span of control. Seven direct reports becomes 20. Twenty becomes 60. The manager who finally has time to coach now has too many people to coach any of them. The mastery crisis and the acceleration addiction are the same phenomenon. Compression creates capacity. Organizations spend the capacity on more output rather than on what the compression freed up.
There is a version of this story where AI finally lets management become what it was always supposed to be. Less administrative work, less status chasing, less translating strategy into task lists. More time for judgment, coaching, culture, and taste. That version is genuinely possible, and it is what the CEO I heard speak was gesturing toward. But it is not the version most organizations are building. The version most organizations are building uses AI's efficiency to justify removing the people who would have done the coaching, widening the span of the ones who remain, and calling the result a flatter, faster, more empowered organization. The rhetoric sounds like liberation. The operational reality is that the conditions for transferring mastery are being systematically worsened.
If mastery transfer used to be incidental, a byproduct of doing the work, then it now must become deliberate. That sounds obvious when you say it out loud, but it runs against the grain of how most organizations operate. Deliberate mastery transfer takes time, and the current business environment is relentlessly hostile to anything that looks like time spent not producing output. So the first move is simply naming it as a real category of work, not a soft skill or a nice-to-have.
The operational practice that goes with this is "show your work." When a manager takes mediocre content and rewrites it themselves, mastery doesn't transfer. The person who submitted the bad version saw that their work was replaced. They did not see what the manager was pattern-matching against, what questions she was asking that they did not know to ask, or what 20 years of reading analyst decks had taught her about why this particular framing would fail. Rewriting the work in silence might be quicker, but it does nothing to produce the next generation of experts. The pattern-matching has to be made visible on purpose.
This is actually where the CEO's advice about pushing conversation into public channels becomes genuinely useful, if you use it the right way. A leader who corrects a PM's work in a public channel and walks through their reasoning is transferring mastery at scale. Five other PMs learn what one PM would have learned in a one-on-one. The public channel is not the problem. Using it for status updates and directives rather than for reasoning-in-public is the problem. The same infrastructure can serve either purpose, and the choice between them is a leadership choice, not a technology choice.
None of this, however, solves the underlying tension. If organizations continue to spend AI's efficiency dividend on expanding span of control, there will not be enough leadership time for deliberate mastery transfer regardless of how good the practices are. That is a real constraint, and I do not want to pretend otherwise. But the leaders I talk to who are wrestling seriously with this have started treating mastery transfer as a design decision, not a hope. They block time for it. They write down what they are pattern-matching against when they give feedback. They do the thinking in public where others can see it. They are "showing their work."
Mastery used to be a byproduct. Now it has to be a design decision. The organizations that figure out the difference will have a capability their competitors cannot build in a hurry. The ones that do not will discover, probably around the time my generation starts retiring, that the capability they assumed would reproduce itself has quietly disappeared.
"Mastery used to be a byproduct. Now it has to be a design decision."
Break a Pencil,
P.S. Know a leader who's trying to figure out how to develop their people in an AI-accelerated organization? Forward this to them.
New to Broken Pencils? Get weekly arguments about product leadership and AI transformation that challenge conventional wisdom. Subscribe here.