When AI does the work, how do employees learn and develop?

Everyone loves a shortcut. A calculator that makes complex math easy. An elevator that saves time getting to the 34th floor. But when we reach for the calculator for simple problems, we lose the ability to estimate in our heads. And if we take the elevator every day, our cardiovascular fitness suffers.

AI shortcuts work the same way, but could have more dire consequences than a potential dip in doing mental math, according to founder of It’s All Relative and generational researcher Dr. Eliza Filby.

Picture a junior employee who is fast, productive, and delivers everything asked of them. They’re likely using AI more effectively than most of their senior colleagues, but over time it becomes clear they've mastered the shortcut, not the material. They can produce a client memo in 20 minutes, but go quiet on a Zoom call because they never really digested the content and therefore can’t answer a client’s follow-up question. Multiply that pattern across a team, and you get a department that moves fast but understands little of what it's producing.

The problem Filby describes isn't laziness. It's a structural gap between what AI-assisted employees can produce and what they actually understand.

Why strong AI-assisted output can hide weak employee skills

Consider the scenario Filby posed to HR leaders at Uplift 2026: when that productive junior employee comes up for promotion, what do you do?

The instinct is to say yes. After all, output matters. But other questions are harder to answer: Has this person developed the judgment to manage others? Can they explain how they got from brief to deliverable? Do they understand the principles behind the work, or just the steps to produce it quickly?

These questions matter because AI has effectively decoupled two things that used to move together: comprehension and production. Before AI-assisted work became standard, a junior employee might spend hours on a memo. Even if the memo wasn't very good, the effort was visible — and a manager could sense whether that person truly grasped the problem or was still finding their footing.

That signal is largely gone now so the only moment skill gaps surface is when the work can't be automated. Like when someone is asked to explain their reasoning on a live conference call, or defend a strategic recommendation to a skeptical executive. Workers sense this, and many avoid those moments by keeping relationships surface-level.

Why performance reviews miss AI-masked skill gaps

Most performance frameworks assess output. That means the manager whose direct report is producing passable work has no systematic way to measure judgment, knowledge depth, and process fluency. To find out, a manager would have to ask, and most haven't been trained or incentivized to have that conversation.

What shadow AI use hides from managers

Filby's research adds another layer to the AI performance conundrum: younger workers are more likely to use unofficial AI tools (also called “shadow AI use”) on personal devices, outside of what their employer has sanctioned. Because that usage can't be tracked, the rates are higher than managers realize, and the skill gaps those tools are masking are also invisible.

There's another type of erosion happening. When employees rely on AI to navigate even mildly uncomfortable situations, they don’t get to practice the kind of human interaction that builds professional judgment over time. Filby spoke about the growing tendency for people to email colleagues two desks away rather than walking over. Similarly, desk lunches have replaced lunch with colleagues, and Slack has replaced hallway conversations. AI-assisted avoidance is accelerating a pattern that was already moving in the wrong direction.

How law firms protect the slow work that builds judgment

Several successful law firms Filby works with have found a deliberate path forward. They keep some of the hard, slow, unrewarded work in place for junior employees. Not as a punishment, but because it's exactly the kind of work senior partners did years ago to develop the judgment and knowledge that makes them good attorneys.

Part of what makes this approach work is transparency about process. When due diligence is taken in asking employees to explain their reasoning and process, not only focusing on the results, then the relationship with AI shifts. They're more deliberate about using it and prepare to stand behind the work it helps them produce. It doesn't require a formal policy, but it does require managers to clarify expectations.

Filby’s research is novel, but the concept isn’t. Confucius is believed to have said: "I hear and I forget. I see and I remember. I do and I understand." The Stoic Epictetus made essentially the same argument — that hardship isn't an obstacle to understanding, it is the understanding. You can't reason your way to capability, you have to be tested by the work.

What that looks like in practice is less about restricting AI than about being deliberate about which tasks still require human struggle. The organizations getting this right have identified the specific experiences that build judgment in their industry, and they protect those experiences even as much of the work around them gets automated.

3 questions HR leaders should ask about AI and skill development

For HR leaders who want to understand where the skill gaps are, three questions are key:

1. Are you monitoring the skills you're trying to develop? Most performance frameworks assess output, which means judgment, knowledge, and process fluency often go unmeasured. Set a baseline for the specific capabilities you're developing in an employee, and watch for moments in coaching or project work where those capabilities get tested.
2. Do your managers know how to assess judgment and knowledge? "Walk me through how you got here" is a different conversation than "your memo looks good." It's also a harder one, and most managers haven't been trained to have it. That gap is worth closing deliberately.
3. Are you accounting for shadow AI use? Younger workers are significantly more likely to use unofficial AI tools outside of what their employer has sanctioned — and that usage can't be tracked. Most AI governance and performance frameworks weren't designed with this pattern in mind. If your visibility into AI use is limited to official tools on company devices, your picture of employee capability may be incomplete.

As AI takes on more of our work, the organizations that succeed won’t be the ones who bought the best AI tools. They’ll be the ones who recognize that AI adoption at scale requires deliberate investment in the capabilities AI can’t replace: judgment, communication, and the kind of knowledge that only comes from doing the actual work.

Read the second article in this series: 3 questions to determine what AI should (and shouldn’t) replace