Most program managers learned their craft by doing the work nobody wanted to do.

Running the tracker. Chasing updates across teams. Assembling the weekly status report from fragments of information that lived in people's heads. Writing the meeting notes that nobody else would write. Manually building the plan, the RAID log, the dependency map — all the artifacts that make a program visible to the people funding it.

It was unglamorous. It was also the best education most of them ever got.

Because through that work, they started to see things. How dependencies actually flow versus how they look on a Gantt chart. Why estimates are always wrong in the same direction. Which teams communicate well and which teams perform well — and why those aren't always the same teams. The admin work gave them the mental model. The repetition gave them the intuition. And over years, that intuition became the foundation for everything that actually matters in this role: people leadership, system design, judgment under uncertainty.

I see this clearly in the PMs I work with. The ones with the sharpest instincts — the ones who can sense a program going sideways before the metrics show it — all built that intuition the same way: through years of accumulating experience — starting with the hands-on, tedious, deeply educational admin work and building from there.

Very soon, AI will be able to do most of it. And as someone who manages a team of PMs, I keep asking myself: how will the next generation learn what the previous generation learned?

The operational work is disappearing

AI is already competent at the tasks that fill a junior program manager's week. Status reports that took hours become a quick review pass. Planning prep that meant half a day of spreadsheet wrangling can be drafted in minutes. Work breakdowns that started on a blank whiteboard can start from a reasonable first draft. Risk registers, estimation sanity checks, dependency tracking — all of it can be accelerated dramatically.

At work, I use Copilot and a sandboxed ChatGPT — reasonable choices for an enterprise that needs to think about data governance and compliance. On my own time, I experiment with tools like Claude Code, n8n pipelines, and agentic workflows that move faster because they don't carry that same overhead. It's not that enterprise tooling is behind — it's optimizing for different things: stability, auditability, control. But even the conservative enterprise tools are already good enough to eliminate a significant chunk of manual PM work.

None of these outputs are production-ready. They need human review, context that only the team has, and judgment about what matters in this specific situation. And when an AI-generated status report gets a fact wrong — a misread dependency, a confidence level that doesn't match reality — the PM is still the one accountable. The verification overhead is real, and anyone who's used these tools knows it's not trivial. But even with that overhead, the manual assembly — the part that used to take most of the time — is increasingly unnecessary.

What the tedious work actually taught us

Here's what I don't think the profession has reckoned with: that administrative layer wasn't just overhead. It was also the apprenticeship.

When a junior PM manually chases down why a team missed a sprint commitment, they don't just get a data point for the tracker. They learn to read human signals — the tech lead who always says "we're on track" until suddenly they don't, the product owner whose silence means disagreement they won't voice in front of the client, the team that delivers late not because they're slow but because they don't trust the requirements.

When they assemble status reports by hand, they develop a feel for which information matters and which is noise. When they build dependency maps manually, they start to see the invisible connections that no tool captures — the ones that run through people, not systems.

That pattern recognition is what eventually allows PMs to grow into the parts of the role that AI can't touch: coaching delivery leads through hard trade-offs, navigating organizational politics, designing how a program should actually run versus how the methodology says it should run, standing in front of leadership when things go sideways and owning the outcome.

The admin work was training data for professional judgment. And we're automating the training data.

The career ladder problem

Among my peers — senior PMs who've been doing this for years — the shift to AI is mostly opportunity. We already have the judgment, the relationships, the instinct for when a program is going sideways before the metrics show it. Automating the admin layer frees us to spend more time on what actually matters: people leadership, system design, the hard conversations that move programs forward.

There's also a mid-level trap that's easy to miss. Many solid PMs have built successful careers on being exceptionally organized administrators — and that's genuinely valuable work. But the shift AI demands isn't just "do the same job faster." It's a different cognitive skill set: systems thinking, business acumen, the ability to operate as a strategic advisor rather than a process operator. Not everyone who's great at the current version of the role will thrive in the next one, and the profession hasn't been honest about that.

But for the next generation, the picture is much less clear.

If a junior PM's first year is spent reviewing AI-generated status reports instead of assembling them by hand, do they develop the same intuition? If they never had to manually track down a blocked dependency, do they learn to see the human dynamics behind it? If the operational work is automated, what's left to teach them?

This isn't hypothetical. I see it already. Junior PMs who are excellent at prompting AI tools to produce artifacts, but who struggle when a planning session goes sideways because they haven't built the pattern recognition to read the room. They can generate a beautiful risk register in minutes. They can't yet sense which risks are real and which are theater.

The profession's dirty secret is that we never had a deliberate training model. We had an accidental one: throw people into the admin work and the good ones absorb the deeper lessons along the way. It worked well enough when the admin work was unavoidable. It falls apart completely when AI makes it optional.

The enterprise gap makes this worse

Here's something that compounds the problem: the speed difference between what individuals can experiment with and what enterprises can responsibly roll out.

I can build an agentic workflow at home that researches a topic, drafts a deliverable, sends it for approval, and publishes it — all autonomously. At work, the AI tooling is more conservative, and for good reason: client data, compliance requirements, audit trails. Enterprises aren't slow because they're clueless — they're slow because the stakes of getting it wrong are higher.

There's a more fundamental issue too: AI is only as useful as the data it has access to. In most organizations, the "source of truth" is fragmented across Slack threads, Jira boards, email chains, and people's heads. AI doesn't fix broken information flows — it accelerates them. A growing part of the PM's job is becoming data orchestration: making sure the inputs are clean, connected, and structured enough for AI to actually deliver value. Without that, you just get confidently wrong summaries faster.

But that creates a real tension for developing PMs. The people who understand AI's potential the best — because they experiment with it personally — develop intuitions that are hard to apply inside enterprise constraints. And the people making decisions about which tools to adopt don't always have that hands-on experience with what the cutting edge feels like in practice. Junior PMs inside enterprises may be doubly disadvantaged: they're losing the apprenticeship and they're working with tools that don't yet show them what's possible.

So what replaces the apprenticeship?

I don't have a clean answer. But I have some instincts from watching this unfold.

Mentorship has to become deliberate, not accidental. Senior PMs can't just assign work and hope the lessons land. We need to explicitly teach the pattern recognition that used to come for free — walk through the why behind decisions, not just the what. "Here's the AI-generated risk register. Now let me tell you which three risks are actually keeping me up at night, and why none of them are on this list."

Structured exposure to the human side needs to start earlier. Facilitation practice. Stakeholder conversations. Conflict navigation. The parts of the job that used to come after years of admin work need to move to the front of the learning path. If the apprenticeship is shorter, we need to compress the curriculum.

And we need to frame AI outputs as conversation starters, not finished products — especially for junior team members. "Here's what the model thinks the dependencies are — what's it missing?" turns out to be a much better learning prompt than "go build the dependency map." It gives people something to react to and surfaces the contextual knowledge that only humans have. The skill becomes critical review rather than assembly, but it still requires deep understanding to do well.

The real bet

I don't think the answer is to resist the automation or to artificially preserve busywork for training purposes. The answer is to experiment — actively, deliberately — with how AI changes the way we work, and to use what we learn to design better paths for the people coming up behind us.

That's what I'm doing on my own time: building AI workflows, testing where they're strong and where they fall apart, developing an intuition for the handoff between human and machine. Not to replace my team, but to understand what the job looks like when the operational layer is handled — and to figure out how to teach that version of the job to someone who never did it the old way.

The program managers who matter in three years won't be the ones who are best at using AI tools. They'll be the ones who can grow other program managers in a world where the old apprenticeship is gone. That means designing new learning paths, creating environments where AI handles the assembly but humans still understand why the assembly matters, and being honest about the fact that the profession's development model needs to be rebuilt from the ground up.

The tedious work made us. Now we need to figure out what makes the next generation — before the profession quietly hollows out from the bottom up.