Yesterday I wrote about the deck that took twenty minutes instead of four hours, and the apprenticeship hiding inside those missing three and a half hours. A few people asked the obvious follow-up: okay, so what do we actually do about it? Here’s what’s being suggested — some of it easier to implement than others, all of it worth thinking through before you decide what fits your team.
Quick Recap: Where the Loop Breaks
Historically, a first-year would get handed something rough — a memo, a first pass at research — and a manager would mark it up. You’d go back and forth. Do that for a few years and you’d eventually be the one doing the marking up. That loop runs on three steps: someone models the thinking, you attempt it yourself, you get feedback on the gap. AI breaks the loop at the attempt stage, because it can now do both the mechanical work and a good chunk of the actual thinking. You get a finished first draft instead of your own flawed one, and you lose the learning that used to come from building something badly and getting corrected.
The idea getting floated as a fix is purposeful apprenticeship: instead of building training around or outside the work, you build it into the work itself, so you keep the speed of AI without losing the learning underneath it. There are four moves worth knowing, and I want to be upfront that some of these are a real culture shift, not just a checkbox.
Move One: Externalize the Reasoning
This is something I’ve worked through with clients directly. Before AI, externalizing your thinking happened automatically, as a byproduct of reviewing someone’s work in progress. Now that AI can do half the work or more, that natural externalization is disappearing. So the fix is deliberate: when a deck comes back from AI, actually write out your commentary — not just corrections, but the praise too. Why a particular slide works, why a certain structure was the right call. If a partner’s only feedback is “looks good,” the junior learns nothing, even when the work is genuinely good. They need to hear why it’s good, or they’re just collecting approvals instead of building judgment.
Move Two: Engineer Friction
This is the one I think is genuinely the hardest to implement. The idea is that after AI generates something — a slide deck, an analysis — you build in a check where the person has to explain back what’s in it. Not a rubber stamp, an actual quiz on their own work product. This is a real culture shift, and I won’t pretend otherwise. But there’s research showing people who hand a problem to AI often can’t answer basic questions about the solution afterward — not because the AI did anything wrong, but because nothing about producing it required them to actually understand it. Play that forward to a client call: someone presents a deck, gets a follow-up question, and has no good answer. That’s the failure mode this friction step is trying to prevent.
Move Three: Gate by Ability, Backwards
This is the move that tends to surprise people. The instinct is to put guardrails around your strugglers and let your strongest people run free. The research says do the opposite. Your fastest, most capable staff are the ones who pick up AI tools quickest — which means they’re also the ones most likely to get very good at producing output while getting very bad at understanding why the tool did what it did. It’s a bit like becoming an excellent programmer who can write flawless code but can’t explain the business decision behind it. That’s a real weakness, and it’s worth catching early, in your best people specifically, not despite them being your best people.
Move Four: Sequence Attempt Before Assist
Have people sketch out their own first pass — even just handwritten notes, even rough and incomplete — before they open AI at all. It’s not always the fastest path in the moment, but it keeps the “attempt” stage of the loop alive instead of skipping straight to a finished draft.
The Honest Tradeoff
None of these moves cost much time on their own — we’re talking seconds to minutes, not hours. If a task used to take four hours and AI gets you a draft in twenty minutes, adding a ten-minute comprehension check still leaves you at thirty minutes total. That’s still an enormous time savings over the old way of working. The real cost isn’t time. It’s that this has to be a manager and partner decision, not something you can hand off to IT or L&D. It has to get baked into how you actually assign and review work, which means it only happens if you’re deliberate about it every time.
The lesson today is straightforward: pick one of these four moves, apply it to one deliverable your team is producing this week, and watch what your junior actually understands by the time it’s done — not just whether the output looks right.
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/how-to-rebuild-talent-pipelines-pt-2/
Leave a Reply