A client of mine recently showed me a proposal deck their first-year associate had put together. It looked great — clean structure, sharp narrative, the kind of deck that used to take a team most of a day to pull together. It took the associate about twenty minutes.
My first reaction was the same as theirs: this is a win. But the more I sat with it, the more a different question started bothering me. That deck used to take four hours. What was actually happening during those four hours that isn’t happening anymore?
The honest answer: a lot more than busywork. Those four hours were the mechanism that turned first-years into managers. And right now, across firms I talk to, that mechanism is quietly disappearing — not because anyone decided to remove it, but because nobody designed it to survive AI in the first place.
Your Review Cycle Isn’t Training Anyone Anymore
Here’s why this matters whether you’re a partner, a manager, or anyone who reviews other people’s work. The review cycle you’ve relied on works like this: a junior drafts something rough, you mark it up, and they learn from the gap between what they wrote and what was correct. That only works if there’s a rough draft underneath it. When AI produces the first draft instead of your staff, your review still happens — but it’s not training anyone anymore. You’re just doing quality control on a tool.
I’ve seen this play out directly with clients who’ve adopted AI heavily for things like proposal decks. If you’ve never spent hours building one yourself, you don’t actually know why it’s structured the way it is. You’re seeing the finished product, not the messy version where you learn why the cost slide goes before the growth story. AI is quietly absorbing the part of the job where that understanding used to get built.
The Apprenticeship Nobody Designed on Purpose
There’s a well-known rule of thumb in corporate learning called the 70-20-10 model: roughly 70% of what people learn at work comes from doing the job itself, 20% from other people, and only 10% from formal training. In public accounting, that 70% has always run through an informal apprenticeship — first-years tying out schedules, drafting workpapers, sitting in the room for the client call before they’re trusted to run one themselves.
Think about how people actually learn to drive a stick shift. Nobody hands you a manual and walks away. You stall the car a few times in an empty parking lot while someone corrects you in real time, and eventually it just lives in your hands. That’s exactly what the grunt work in accounting was doing. The first draft of a memo wasn’t valuable because it was good — it usually wasn’t. It was valuable because producing it badly and getting corrected is how judgment actually gets built.
Why This Time Is Different
Here’s what makes this round of automation different. The calculator took over arithmetic but left number sense alone — staff still had to know when to divide and whether the answer looked plausible. AI doesn’t sit underneath the judgment layer — it reaches directly into it, structuring the argument, weighing the position, drafting the actual recommendation a client acts on. Same activity, two very different jobs happening at once. You can’t keep one while losing the other.
McKinsey’s Chief Learning Officer has talked publicly about watching this erosion happen at her own firm, especially after COVID scattered teams. Her team’s conclusion was blunt: you can’t hope the old apprenticeship comes back on its own. You have to design it back in — what she calls purposeful apprenticeship.
What This Means for Your Engagements
So what does this mean day to day? Your review still catches errors, but it’s no longer building anyone’s skill, because there’s no rough attempt underneath the polished output to compare against. Research on self-explanation backs this up — people who explain their own reasoning, even badly, catch far more of their own errors than people who only review someone else’s finished version. Lose the attempt, and you’ll find out three years from now when a senior can’t defend a position under client pushback, because they never actually built the reasoning themselves.
And here’s the part that surprises people: it’s often your best performers who are most exposed, not your weakest. Staff who are still struggling will keep attempting things manually because they need the scaffolding. Your sharpest associates are usually the ones efficient enough to let AI handle the entire task from day one — which means they’re the ones most likely to never build the underlying judgment at all.
Key Takeaways
- The 70-20-10 split that built your current managers depended on juniors doing real work badly and getting corrected. That work hasn’t disappeared — it’s hidden inside an AI output now.
- AI is different from past automation because it reaches into the judgment layer, not just the mechanical layer underneath it. The task and the training were always the same activity.
- Your review cycle can still catch errors without building anyone’s skill, if there’s no rough attempt underneath the polished version to compare against.
- Don’t assume your top performers are fine just because they’re fast — they may be most likely to skip the rep that builds real judgment.
Tomorrow, I’ll walk through exactly how firms are starting to rebuild this on purpose — specific changes to how work gets assigned so the next AI-assisted deliverable your team produces actually builds judgment instead of just looking finished.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: https://everydaycpe.com/courses/non-technical-personnel-human-resources/ai-is-breaking-talent-pipelines-pt-1/
Leave a Reply