Learning & AI

Imagine two new analysts, both smart, both eager. Call them Sarah and David. On their first day you hand them the same powerful AI tool. It can read a thousand pages of dense financial reports and produce a clean executive summary in thirty seconds.

Sarah uses it as a partner. She writes her own summary first, struggling through the jargon and the numbers, then runs the AI’s version and compares. She looks for what she missed, where it found a connection she didn’t. She argues with it, rephrases its sterile prose, uses the contrast to sharpen her own thinking. The work is still hard. It just moves faster.

David uses it as a vending machine. He feeds in the documents, gets a summary, it looks good, he ships it. His manager is happy with the turnaround. He does this again and again.

Six months pass.

Sarah is now one of your sharpest thinkers. She walks into a meeting having read the pre-brief and holds her own with the veterans. She has internalized the patterns of good analysis and rarely needs the AI for the core task anymore, though she still uses it to check for blind spots. She is dramatically more capable.

David can no longer do his job without the tool. If the wifi goes down he’s useless. He has gotten very good at writing prompts and no better at analysis. His output is high, his actual skill a flat line. He’s a high-speed operator of a black box.

This is the augmentation gap, the widening chasm between people who use AI to get better and people who use it to avoid the work of getting better. As a leader, the most important question about AI is no longer “should we use it,” but which side of this gap your team is on.

The story we’re told about AI at work is a story about productivity. Bolt it on, the numbers go up. Reports get written faster, code gets generated instantly. On a spreadsheet this looks like an unqualified win.

But productivity is not capability. A team that is only productive, not capable, is brittle. A fast car with a tiny gas tank, running on fumes.

The myth is that the work is the output. The real work, the work that builds skill, is the struggle: the frustrating, messy process of wrestling with a problem until you understand it. Every time you remove a rep, you remove a learning opportunity. The AI that gave David the summary also took from him the chance to learn how to synthesize one himself.

Call it the paradox of practice. The very things that make a task difficult are often the things that make it valuable for learning. A calculator is a wonderful tool for an engineer and a disastrous one for a third-grader still learning multiplication. Hand it over too early and you don’t just help them get the answer, you prevent them from ever learning the method. We are now handing out calculators for everything, and calling it progress.

This doesn’t make AI the enemy. It means we need a better frame. The question isn’t if your team uses AI, it’s how. Does it accelerate learning, or enable coasting?

Two ideas help. Learning velocity is the speed at which a person or team converts new information into durable, usable capability. Sarah had high learning velocity: the AI gave her instant feedback, a tireless sparring partner, more high-quality reps in less time. Training half-life is the rate at which a skill decays without practice. David is experiencing a catastrophic shortening of his, because he never practices the core skill, so his ability to do it atrophies and he becomes a human wrapper for an API call.

Used correctly, AI is the single greatest tool we have ever invented for raising learning velocity. Used incorrectly, it is a systematic way to hollow out the capability of an entire organization, one task at a time.

This isn’t a philosophical point. It’s grounded in biology.

Before I built companies, I was a scientist. I spent my bioengineering PhD putting people in brain scanners, trying to understand how networks in the brain learn and change. The most reliable finding in all of neuroscience is this: learning is not a download. It is a physical process of building and strengthening connections between neurons.

What drives it? Effortful retrieval. The act of trying to pull information out of your own head is the signal that tells the brain this is important, keep it, make it faster next time. When you watch an answer appear on a screen, none of that happens. The struggle is not a bug. It is the entire feature.

After the PhD I spent over a decade building AI in the real world. My first company, RoadBotics, used machine vision to help governments decide which roads to fix, and we sold it to Michelin. The throughline of my whole career is one belief: technology should augment human intelligence, not replace it. It should make people more capable, more insightful, harder to replace. An AI that just gives you the answer makes you easier to replace. An AI that helps you learn to find the answer yourself makes you indispensable.

Your job has changed. You are no longer just a manager of tasks and timelines, you are the curator of your team’s reps. Before you roll out any AI tool, before you automate any process, ask one question:

Which reps does this remove?

And the follow-up: are those the reps that build the skill we actually need?

Sometimes the answer is easy. Automating the copy-paste of data between spreadsheets removes a rep that builds zero valuable skill. Let the machine do the toil. But an AI that writes the first draft of every email for your new sales rep removes the reps of thinking about the customer, finding a hook, structuring an argument, the very practice that turns a rookie into a closer. You will see a short-term bump in email volume while you quietly destroy your talent pipeline.

Keep the human in the loop for the hard parts: the judgment, the synthesis, the creative leap, the empathetic connection. Let AI handle the rest. Let it be the research assistant, the brainstorming partner, the tireless simulator. Never let it be the thing that does the thinking.

This is not a solved problem. We are all early in a massive, uncontrolled experiment on human capability, and the augmentation gap widens every day, in every company. On one side, teams compounding their skills at a dizzying rate. On the other, people who look productive on a chart but are one software update from obsolescence. Which side your team lands on is being decided right now, in the small choices you make about the tools you adopt and the workflows you design, whether you are paying attention or not.

I write about this most days on LinkedIn, figuring it out in public. The essays below go deeper.

• The augmentation gap
• Learning velocity
• Training half-life
• Team Capability
• Leading Through Change