There's a cost to letting AI do the work that nobody puts in the brochure. It's not about privacy or accuracy or the usual list of concerns. It's subtler than any of those things.
It's about memory.
When you stop doing tasks, you stop remembering how to do them. And memory, it turns out, isn't just something you have — it's something you build, through the physical and cognitive act of doing the work yourself.
Why Your Own Notes Were Always Better
Think back to school. You know that feeling when a classmate shared their notes with you — neatly organized, legibly written, probably better than yours? You read them. You understood them. And then, two days before the exam, the information was just... gone.
Compare that to the notes you took yourself. Messy, abbreviated, written in a hurry. You had to wrestle with the material to get it onto the page. But a week later, you still had it. You could reconstruct the argument without looking. You owned it.
Psychologists have studied this since the 1970s under the name the Generation Effect. Information that you produce yourself — that you struggle to articulate, encode, phrase — sticks in memory far more securely than information you passively receive. When you write the code, you're not just producing an output. You're building a neural pathway. When you draft the essay, you're not just filling a document. You're practicing reasoning under your own name.
When AI does it and you review the result? You're reading someone else's notes.
Search Was a Warm-Up
We got a preview of this with search engines. The "Google Effect" documented how people stopped retaining facts once they knew they could just look them up — and honestly, that turned out fine. You don't need to memorize the population of Uruguay. Offloading trivia is a reasonable trade.
But AI is doing something different and considerably more significant. It doesn't just store facts. It performs processes.
There's a meaningful difference between knowing you can look up a capital city and knowing that AI can structure your argument, debug your code, or plan your project. Outsourcing data is one thing. Outsourcing the struggle — the iteration, the dead ends, the trying-and-failing — is something else entirely, because the struggle is precisely what builds expertise.
Expertise isn't a database of correct answers. It's a network of experiences, most of them bad ones. A senior engineer knows how to fix a broken system because they've broken similar systems before and had to live with it. A good writer knows what works because they've written terrible drafts and had to figure out why they didn't land. You can't download that. It gets built through contact with the work over time.
If AI handles the drafting, the debugging, and the structuring, you get the output — but you skip the thing that was actually making you better. You produce good work in the short term while quietly stopping to grow.
The Shallowing
For now, this is manageable. Most AI use is still task-by-task, and the tasks being delegated are often peripheral to someone's core expertise. But agents are getting better and the scope of what they can handle is expanding. As that happens, the pattern compounds.
The more you delegate, the less deep experience you accumulate. And over time, this creates a risk I think of as shallowing — where you become someone who knows enough to prompt the AI, but not enough to do the work themselves. You have broad awareness across many areas, but your depth in any of them starts to thin. You're a coordinator, not a practitioner.
That's a real trade-off. Not a hypothetical one.
Being Honest About the Trade
I'm not arguing we should make work artificially hard or pretend AI isn't useful. Drudgery is drudgery. If AI can do the tedious part of your job while you focus on the parts that actually matter, that's a genuine improvement and you should take it.
But "automate everything by default" is a choice, and it has consequences that are easy to ignore because they accrue slowly. The question worth sitting with isn't whether AI is capable of doing the task. It's whether doing the task yourself is actually building something in you — and whether you'd miss that if it were gone.
Are you offloading this because it's a waste of your time, or because it's hard and you'd rather not do it? Do you want to be an expert in this area, or just a manager of it? Is this a skill you care about developing, or one you're genuinely fine handing off?
There's no universal answer. Delegating email? Go for it. Delegating the core reasoning of your profession, the thing that makes you valuable and keeps you sharp? Be careful with that one.
The future of work might look very productive on the surface — polished outputs, fast turnaround, clean results. But underneath that efficiency, it's possible to quietly forget how to do the very things that got you there. The notebook you wrote yourself, even messily, is still the one you remember.