Your AI Rollout Might Be Quietly Breaking Your Talent Pipeline

By Kamran Akbar · August 26, 2026 · 4 min read

A senior businesswoman and a younger colleague reviewing documents together at a desk in a modern office

Key takeaways

A Goldman Sachs partner warned on August 24 that AI use risks "cognitive atrophy" in junior bankers who stop reasoning from first principles.
The bigger risk isn't AI getting things wrong, it's removing the apprenticeship work that used to build judgment in new hires.
Small businesses have less room to absorb this than Goldman does, since they often lack a deep bench of senior staff to catch AI mistakes.
Roughly two thirds of workers already worry AI is eroding their own writing and creative abilities, according to research cited alongside this story.
The fix is structural, keep a human decision point on consequential work and rotate junior staff through manual reps even after a task gets automated.

Chris Churchman spends his days pushing Goldman Sachs deeper into AI. He leads the bank's Marquee platform and co-chairs its Global Banking and Markets AI working group, and that platform just helped drive 39 percent year over year revenue growth in the second quarter. So it was notable when, on a Goldman Exchanges podcast on August 24, he stopped selling AI for a moment and warned about it instead.

His words were blunt.

"There's a huge danger here that in the era of AI, we outsource our reasoning to these models, and we have cognitive atrophy that stops us being able to reason from first principles ourselves."

Even more striking, Churchman admitted Goldman "hasn't yet figured out" how to manage that tradeoff, despite running the bank's own AI strategy. That admission is worth sitting with, because if the firm with the resources, the data, and the incentive to get this right is still stuck, most other businesses have not even started thinking about it.

The real risk is not the model, it's the apprenticeship

Churchman's specific worry is not that AI gets things wrong. It's what happens when junior bankers stop doing the grinding, repetitive work that used to be how they learned the job. Building a financial model from scratch, structuring an argument for a client pitch, working through a pricing request under the eye of a senior trader, none of that work was really about the output. It was the training ground where judgment got built, one rep at a time.

Automate that layer of work and you don't just save time. You remove the apprenticeship. Churchman's fear is that Goldman ends up with a generation of what he half joked are "prompt engineers" rather than bankers who can reason from first principles the moment a model is wrong, silent, or simply not in the room.

This is not only a banking problem. Coverage of Churchman's comments in HR and learning and development circles pointed to research showing roughly two thirds of workers already worry AI is eroding their writing and creative abilities. One recommendation raised there, from professor Mohammad Hossein Jarrahi, was structural rather than motivational. Build workflows that require employees to actively challenge AI output, instead of workflows that just ask them to approve it.

Why this matters more for a small business than for Goldman

Goldman can absorb a rough decade of undertrained mid-level bankers. Most small and mid-sized businesses cannot. If you run a ten, fifty, or two hundred person company, the entry-level work you're tempted to hand fully to AI right now, first-draft contracts, first-pass financial models, first-round customer research, junior-level code, is very often the exact work that used to turn a new hire into someone you could trust with judgment calls two years later.

Cut out that rung and you save money this quarter. You may also discover, a few years from now, that nobody on your team can catch the AI when it's confidently wrong, because nobody ever had to learn how to do the work without it. That is not a hypothetical for a company with ten people. It's an existential one, because you likely don't have five other senior people to catch the mistake for you.

What to actually do about it

Churchman offered no polished playbook, and honestly nobody else has one yet either. A few practical moves are already showing up at companies taking this seriously.

Keep a human decision point on anything consequential. Automating the first draft is fine. Automating the sign-off is how you lose the skill for good.

Rotate junior staff through the manual version of a task on a regular schedule, even after it's automated, the same way pilots still log manual landings so the skill doesn't disappear.

Make "explain why the AI got this wrong" part of onboarding and periodic review, not just "use the AI tool correctly."

Track which roles are quietly becoming pure AI-output reviewers with zero hands-on reps, and treat that as a retention and skills risk, not just an efficiency win.

None of this argues for slowing down AI adoption. It argues for treating your training pipeline as infrastructure, the kind that quietly disappears if nobody is deliberately protecting it, and the kind you will not notice is gone until the day you need it most.

My take on AI and the junior talent pipeline

When I read Chris Churchman's comments about Goldman Sachs not having this figured out, my first reaction was relief, honestly. If a bank with that much money and that many smart people is still working through it, none of us should feel behind.

My actual worry isn't about the big models getting something wrong on a Tuesday. It's about what happens two or three years from now when the people I hired straight out of school never had to do the boring, repetitive version of the work. That boring version is where they would have learned to spot a bad number, push back on a client's assumption, or catch a mistake before it became expensive.

In my own team, I've started treating certain tasks as training reps that stay manual on purpose, even when a tool could do them faster. A first-pass budget, a first draft of a contract clause, a first attempt at diagnosing why a customer churned, I want a person to sweat through those a few times before they're allowed to just review AI output and sign off.

That's slower this quarter. I think it's the only way I end up with people in three years who can actually catch the AI when it's wrong, instead of a team that's very good at approving things quickly and has no idea why.

Questions people ask

What exactly did the Goldman Sachs partner warn about?

Chris Churchman, who leads Goldman's Marquee platform, said on a Goldman Exchanges podcast on August 24, 2026 that heavy reliance on AI risks causing cognitive atrophy in junior bankers, meaning they lose the ability to reason from first principles because the model is doing that reasoning for them.

Why does this matter for a small business, not just a big bank?

Goldman has the resources and depth of staff to absorb a few years of undertrained junior employees. Most small and mid-sized businesses do not, so if AI removes the entry-level work that used to train judgment, there may be no one left who can catch a mistake the AI makes.

What can a business actually do about this?

Keep a human decision point on consequential work rather than automating the sign-off, rotate junior staff through manual versions of a task on a schedule even after it's automated, and build review habits that require employees to explain why an AI output might be wrong rather than simply approving it.

Does this mean businesses should slow down AI adoption?

No. The point isn't to avoid AI, it's to treat the training pipeline that builds future judgment as something you have to protect on purpose, the same way you would protect any other asset that can quietly disappear if nobody is watching it.

Sources

Want help putting this into practice?

Let's talk