I’ll say this from my own time at PwC: the reviews I actually remember, years later, weren’t the ones where a partner just fixed my numbers and moved on. They were the ones where a partner stopped, said “here’s why this is wrong,” and made me sit with the reasoning before we went any further.
That habit is exactly what’s at risk right now — and the reason has nothing to do with training budgets or firm policy.
A Number You’ve Heard, Colliding With One You Haven’t
Here’s the number you’ve heard: 75% of CPAs are retirement-eligible within the next 15 years, according to AICPA data. Firms have known about that cliff for a while.
Here’s the one you probably haven’t connected to it: the mechanism firms used to rely on to capture a partner’s judgment before they walked out the door is quietly disappearing right now — for a completely unrelated reason. AI drafts the first pass. Reviewers skim instead of annotate. And the two trends are hitting at exactly the same time.
The Accidental Knowledge Pipe
The old apprenticeship model ran on a byproduct nobody designed on purpose. A partner would mark up a junior’s draft, and in explaining why something was wrong, they were forced to say out loud — or write down — reasoning they’d otherwise never articulate. Researchers call this externalization: turning knowledge that lives in someone’s head into knowledge that lives somewhere else. It’s the hardest kind of knowledge to move, and for decades, the review process moved it without anyone calling it that.
Think about it like an old factory that vents its own exhaust into a greenhouse next door. Nobody built the pipe on purpose — it just turned out that running the plant kept something else alive. Review markup was that pipe. Partners weren’t trying to train the next generation every time they marked up a draft. They were just doing their job. Training happened as a side effect.
Here’s what that looked like in practice. Picture a senior manager reviewing a revenue recognition memo. The junior got the conclusion right, but cited the wrong contract clause to get there. A good reviewer doesn’t just swap in the correct clause number and move on — they stop and say, “here’s why that clause doesn’t apply. Look at the termination provision in section four. That’s what actually controls.” That thirty-second detour is the entire apprenticeship model in miniature. The junior didn’t just get a correct memo. They got a transferable piece of judgment they’ll reuse on the next five contracts that cross their desk.
Why the Pipe Is Closing
When a junior hands you their own rough first draft, your markup teaches real problem-solving — you’re reacting to their actual reasoning, mistakes included. When AI drafts the memo and the junior passes it through, your read shifts. You’re not asking “is the reasoning sound” anymore. You’re asking “does this look right.” That’s a lighter, faster review — and it doesn’t produce the same byproduct. The partner’s judgment stops getting written down at exactly the moment the retirement clock is loudest.
Here’s how it plays out badly if nobody catches it: a senior manager leaves, and nobody worries about it because the team’s numbers looked fine for years. Then eighteen months later, a new hire runs into a judgment call the manager used to make instinctively — how to treat a borderline related-party disclosure, say — and there’s no one left who remembers the reasoning, only the historical treatment. That’s the expert blind spot compounding upward. The gap doesn’t show up when the expert leaves. It shows up quietly, months later, on the first edge case nobody wrote down an answer for.
Three Things That Cost You Nothing to Start
None of this requires new software or a training program. It requires a reviewer deciding to do three things differently, starting with the next file on their desk:
- Ask before you grade. When someone hands you AI-assisted work, don’t lead with your verdict — ask them to walk you through how they got there. It works whether AI or a human drafted the first version, and it costs nothing extra.
- Write the why, not just the fix. “Change this number” teaches nothing. “Change this number — because the client’s fiscal year-end doesn’t match the standard cutoff we assume” teaches the thing that used to transfer for free.
- Spot-check without AI, occasionally. Not as punishment, not on everything — just enough on a real file to see whether the judgment is actually there underneath the polished output, before it matters on a real client issue.
Key Takeaways
- 75% of CPAs are retirement-eligible within 15 years — the tacit knowledge clock is loud.
- The old apprenticeship model captured judgment as a byproduct of reviewing rough, human-made drafts.
- AI-drafted first passes make review faster and lighter — and stop producing that byproduct.
- The fix isn’t a program. It’s asking “walk me through how you got here,” writing the one-sentence why, and occasionally spot-checking judgment without AI.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]
Leave a Reply