I want to tell you about two people I'll call S and M, because the difference between them is the most important thing I've watched happen this year, and I think it's happening in your life too, somewhere, even if you haven't named it yet.
They work at the same company. Same team. Same access to the same AI tools. Hired three months apart by the same manager. Both articulate. Both, by every standard performance metric, doing well.
S, every morning, opens her notebook before her laptop. Fifteen minutes with a pen on whatever's hardest that day. A rough framework. Gaps she can name. A direction she can defend. Then she opens the AI tools. She uses them like a surgeon uses instruments — to challenge what she's already thought, to find the cracks, to surface alternatives she hadn't considered. The understanding stays hers. The tool is the instrument.
M opens the AI tools first.
He's better with them than S, in raw prompting skill. His deliverables are polished, comprehensive, often excellent. Clients are happy. His manager has no specific complaints.
Six months in, S is the one anyone wants in the room when the meeting goes sideways. M hedges when asked for a view that wasn't in the deck.
Nothing is wrong with M. He isn't lazy. He's doing what his environment rewards — output speed, polish, comprehensiveness — in the way the environment taught him to do it.
The catch is small and slow and devastating: *excellent outputs do not accumulate into expertise.*
Expertise is built through the work of developing understanding. When the work is delegated, the expertise does not develop. The person produces without building. And the gap between producing and building widens every year, almost invisibly, until it isn't invisible anymore.
I want to be careful here. This is not a story about who uses AI and who doesn't. Both S and M use AI extensively. The tools are identical. What's different is the relationship to the tool.
S uses AI as an amplifier of thinking she has already done. M uses AI as a substitute for thinking he hasn't yet done.
Stated like that, the distinction sounds simple. Even obvious. The cost of the difference, however, compounds over years in ways no single observation can capture.
Every time S thinks first and uses the tool second, she builds something. A small increment of understanding. A slightly richer model of the domain. A more nuanced grasp of how the factors interact. Over months and years, those increments accumulate into something we used to be able to name without irony: real expertise. The kind that lets you navigate the unexpected, evaluate the tool's output, adapt when the situation changes — be the person in the room who actually understands what's happening.
Every time M skips to the tool, he produces something. An output. Often excellent. The output does not accumulate. The output is gone the moment the next task arrives.
In six months, the gap between S and M is already present in the quality of their judgment under pressure. In two years, it will be visible to anyone who knows what to look for. In five, it will be undeniable.
The divergence is happening already.
Three forces push almost everyone toward the M path, and I want to name them, because the names are how you begin to resist them.
The first is economic pressure. Most professional environments measure outputs immediately and measure underlying understanding not at all. The choice to think first imposes a cost — slower, harder, less polished initial draft. The reward — capability that holds up when the tool isn't there — is delayed by years and never appears on a performance review.
The second is social normalization. Heavy AI use isn't merely accepted in most knowledge work; it's expected. The person doing substantial cognitive work without AI assistance is, in many environments, viewed as inefficient or eccentric. Social pressure does not yield to long-term arguments. It is one of the more reliable behavioral forces we know.
The third is feedback latency. Every time you use the tool and get a good output, you receive immediate positive reinforcement for the behavior of using the tool. The downside — atrophy of the underlying judgment — is delayed by years and never surfaces as a discrete signal. There is no notification that says your analytical capacity declined slightly today because you delegated reasoning you should have done yourself.
The reason this matters more for the people reading this letter than for most other populations is uncomfortable but true: this divergence is happening fastest among knowledge workers. Among the people paid for the quality of their judgment. Among the people who, in five years, will look around and find that some peers have compounded a kind of capability the rest of us cannot easily catch up to.
If you've felt the pull toward the M path — toward skipping the thinking because the tool is right there and almost always pays off — that's not a moral failure. It's a perfectly rational response to your environment. The environment is the problem. Knowing the mechanism does not free you from it. But it lets you, occasionally, do something the environment doesn't reward: think first, use the tool second, build something that compounds.
This is not advice. It's just what I keep noticing.
The split future already started. The gap is small now. It's still small enough to close.