AI and Talent Development: What Gets Lost

Ai and talent development: What Happens to the Judgment When the Work That Taught It Disappears

The savings are real. That part isn’t in dispute. A finance function that once required a dozen analysts to produce the monthly close, the variance analysis, and the scenario models can now do much of that work with a fraction of the headcount and a set of AI tools. The senior people are still there. The outputs still land. The cost structure is genuinely different.

What’s harder to see, because it won’t become visible for several years, is what the organization also changed when it eliminated most of the analyst layer.

It didn’t just cut a cost center. It removed a development curriculum.

AI and Talent Development: What Junior Work Actually Teaches

There’s a version of analytical work at the entry level that looks, from the outside, like execution. The analyst builds the model, updates the dashboard, runs the numbers, formats the slide. Efficient in, efficient out.

What that description misses is what happens inside the analyst while they’re doing it.

They learn where the data is unreliable, not in the abstract, but specifically, in this company’s systems, this source tends to overcount in Q4, this field gets updated inconsistently. They learn to read a spreadsheet for the story inside it rather than the numbers on top of it. They learn what a senior person actually needs versus what they literally asked for.

They make mistakes. They produce an analysis that misses something obvious, someone catches it, and someone explains why. Over time, they develop the judgment to catch it themselves before it goes upstream. They learn what confident wrongness looks like — a number that seems clean but isn’t — and how to build a habit of checking before it becomes a habit of trusting.

Call it what it actually is: source skepticism, anomaly recognition, the ability to reconstruct a process backward from a suspicious output, calibrated confidence about when something is probably fine and when it needs a second look. None of this appears in a job description. All of it is the job.

The mechanism underneath it is a loop: produce something, get it wrong in a specific and traceable way, receive correction, understand why, try again. Judgment accumulates through that loop, not through the task itself.

What the Senior Person Learned That’s Harder to Replicate Now

The VP who now manages a much smaller team came up through a version of that loop, repeated for years, across an analyst role, then a senior analyst role, then a management role overseeing analysts. None of it was designed as formal training. It was just how the work got done, and the loop ran continuously underneath it.

That progression gave the VP something specific: they know what’s inside the process. When a model produces an output that looks right but isn’t, they recognize the shape of the error before they can fully explain it. They know which assumptions deserve pressure-testing and which inputs are usually clean. They learned this by being wrong inside the process, repeatedly, with feedback.

The people now on their team have a different set of experiences. They’ve reviewed AI-generated outputs. They’ve gotten skilled at prompting, spotting obvious errors, managing exception cases. That’s not a lesser skill set — it may turn out to be an equally important one. But it’s a different loop, and it’s not yet clear it produces the same depth of process-level judgment the old one did.

This isn’t a deficiency in the people. It’s a question about the pipeline, and it belongs to the organization, not to them.

Automation and Augmentation Are Not the Same Thing

The distinction that matters here isn’t whether AI is involved. It’s whether AI removed the junior person from the feedback loop or changed their position inside it.

If the workflow becomes prompt → receive output → approve, the loop that used to build judgment has effectively disappeared. The junior person is now a checkpoint, not a participant in the error-correction cycle that used to shape how they think.

If the workflow instead becomes prompt → inspect → verify → challenge → explain → revise, something different happens. The junior person is still exposed to error, still forced to reconstruct reasoning, still developing the skepticism that catches a plausible-looking wrong answer. That loop may preserve the developmental function, possibly even accelerate it, since AI can generate far more edge cases and scenarios to practice judgment against than a slower manual workflow ever could.

Early evidence on this is suggestive rather than settled. Stanford researchers have documented disproportionate employment declines among early-career workers in occupations most exposed to generative AI, particularly where the technology looks more like automation than augmentation, and separate ongoing research is directly examining whether AI expands what apprentices can produce at the cost of what they actually understand. Neither line of research proves that judgment formation is collapsing. Both suggest the question deserves more attention than it’s getting.

The organizational failure isn’t using AI. It’s using it in a way that quietly removes people from the loop that used to teach them something, without noticing that’s what happened.

The Succession Question That’s Easy to Miss

Organizations that automated their analyst functions largely made those decisions at the function level, looking at current operations. The CFO or business unit head assessed what AI could now do, weighed the cost savings, and made a reasonable near-term call.

The question that’s easy to miss in that calculation is what the pipeline looks like several years out, when the VP eventually moves on, and the organization discovers who’s actually positioned to replace them.

That’s not a prediction that the successor will fail. It’s an open question worth asking now, while there’s still time to do something about the answer: has this person developed the specific judgment — the ability to sense when a clean-looking output is wrong, to work backward through a process they’ve never had to build from scratch — that the previous generation developed by default, just by doing the work?

Senior capability doesn’t come from task repetition alone. It also comes from coaching, rotations, expanding responsibility, decision exposure, mistakes made at increasing altitude. Automating entry-level tasks doesn’t eliminate all of that. It removes one developmental pathway, a significant one, without necessarily replacing what it used to produce.

If organizations remove that pathway without deliberately redesigning it, they may find, later, that their bench contains titles without the experience that used to come attached to them.

The Pattern Beyond Finance

Finance is the clearest example because the loop is easy to see, but the question isn’t limited to it.

Legal is a useful second case, because the profession is already discussing it explicitly. As AI absorbs more first-pass contract review, legal research, and due diligence, commentary in the field has increasingly framed this as a training problem rather than only an efficiency one, arguing the answer isn’t preserving busywork for its own sake, but deliberately rebuilding junior legal training around verification, judgment, and the ability to catch what the AI misses.

The same underlying question — what did this work develop in people, and where does that development happen now — applies wherever entry-level analytical or drafting work is being absorbed by AI tools. It doesn’t need a separate case study in every function to make the point.

What’s Worth Asking Before the Pipeline Question Becomes a Crisis

There’s an honest complication worth naming: a lot of old junior work really was repetitive, low-value, and exhausting. Nobody should want to preserve inefficient manual tasks purely because they happened to double as training. That would be nostalgia, not strategy.

The better question, before automating a layer of work, has two parts: what output does this work produce, and what capability does doing it produce in the person doing it? If a task disappears and the answer to the second question was significant, the developmental function that used to ride along with the task has to become explicit somewhere else — deliberately designed, not assumed to survive on its own.

That’s harder than preserving headcount. It also might be the only real solution.

There’s a market-level version of this problem too, worth sitting with for a moment: every company wants to eventually hire experienced senior leaders. But the market can only supply people with that experience if enough companies were willing, somewhere along the way, to pay for the slower, less efficient process of developing it. If every organization automates its own version of this pipeline simultaneously, the assumption that experienced talent will simply be available later stops holding.

The savings are real. So is the question of what they’re quietly trading away — and whether anyone in the organization is actually accountable for finding out.

Charlie Solorzano is Managing Partner at Alder Koten, an executive search firm focused on C-suite and board leadership across the U.S. and Mexico. He advises founders, investors, family businesses and boards on executive search in Mexico, U.S.–Mexico cross-border leadership, succession and leadership transitions. His work is based by The Race Conditions Model™, which examines the environment an executive will enter, and The Driver Calibration™, which examines how that executive operates within those conditions.