The Irreplaceability of Reps
You can’t become a surgeon by watching surgery videos.
You need to practice. Doing the small, boring things over and over again until they become second nature and you move on to other small things.
Repetition is what separates the great from the average.
As Ryan Holiday writes, “All success is a lagging indicator” of the work you’ve put in.
“Your retirement accounts are a lagging indicator of whether or not you have your financial act together… Hitting a personal record on the bench press is a lagging indicator of a lot of discipline and hard work.”
Robert Greene says, “creativity is a function of the previous work you put in.”
Every single big thing out there is a culmination of the repetitive, boring, grunt work being done over and over again.
Because the grunt work helps you build:
- Tacit knowledge (the stuff that can’t be written down)
- Error recognition (you learn what “good” looks like by producing “bad”)
- Judgment under uncertainty (the ability to make calls when the answer isn’t obvious)
- Grit/tolerance for frustration (the psychological muscle to push through difficulty)
Over time, this compounds. Year 1 makes Year 2 easier. Year 2 makes Year 5 possible. Skip Year 1, and Year 5 never comes.
The Shortcut Temptation
AI offers to do the grunt work for us: the reports, the debugging, the research.
At first sight, this sounds like pure upside. Why waste time on repetitive tasks when AI can do them?
But the reps are the point. You don’t make 1000 PowerPoints to make PowerPoints. You do them to learn:
- Which information matters
- How to work under pressure
- How to spot inconsistencies in data
- How to tailor the PowerPoint to different audiences
The Skill Hollowing
If AI does entry-level work, two things happen:
- No one learns the fundamentals (you can't run if you never learn to walk)
- A gap starts forming in the system:
Once entry-level work is automated, intermediate level becomes the new “entry point.”
But intermediate work assumes you did the entry-level grind. We all know a middle manager who doesn’t know anything about the job and tries to tell people what to do although they have no idea themselves. Now imagine that’s everyone.
We'll become managers of processes we do not understand.
People who delegate to AI but can’t verify the output. That’s what we’re building.
Now scale that across an entire organization. An entire industry.
If no one did the foundational work, no one can check if the AI is right. The developer can’t catch security flaws. The analyst can’t spot flawed assumptions. The lawyer can’t verify case citations.
We end up in a world of:
- Plausible-sounding nonsense (AI is great at this)
- Undetected errors (because no one has the competence to spot them)
- Systemic fragility (one AI failure cascades because no humans can take over)
The Power Concentration
But the people who currently do the hard work, today’s experts, AI company founders, senior practitioners…they keep their competence. Everyone entering the workforce becomes dependent on their tools.
I'm not writing this to bring you in on some huge conspiracy. It’s just what happens when one group has the skills and another group doesn’t. Competence asymmetry becomes power asymmetry.
The Hard Work Isn’t Optional
AI is amazing for augmenting competence but it can’t create competence from nothing. If we use it to skip the grind, the reps, we don’t get to keep the expertise.
If we don’t keep the expertise, how do we operate?
The reps are a feature, not a bug.
The Counterargument
You might think: “Okay but won’t people notice they’re becoming less competent? Won’t they realize AI is making them weaker and self-correct?”
No. The degradation is gradual and invisible.
When you use AI to do entry-level work, you don’t suddenly become incompetent. You just never become competent in the first place.
Worse: AI makes you feel more competent than you are. You can produce output that looks professional. The reports get written. The code runs (mostly). The presentations get delivered. You get positive feedback. You might even get promoted.
The gap only becomes visible in two ways…
First: You hit a problem the AI can’t solve. A novel situation, an edge case, a judgment call that requires deep understanding. The tool breaks down or says “I can’t help with that” and you realize you have no idea how to do this yourself.
Second (worse): The AI doesn’t break down. It confidently gives you an answer. The answer is wrong, but it sounds right, it’s plausible. And you have no way to tell it’s wrong because you never developed the expertise to evaluate it.
The code compiles but has a subtle security flaw. The financial model looks sophisticated but uses flawed assumptions. The medical diagnosis sounds reasonable but misses a critical contraindication. The legal brief cites cases that don’t actually support the argument.
And nobody catches it.
By then, we won’t just lack the skills. We will also lack the metacognitive ability to recognize that we lack the skill. We won’t be able to tell good work from bad because we never did enough bad work to learn the difference.
This is the real danger: not that people will knowingly choose incompetence, but that they’ll unknowingly choose dependency and mistake it for competence.
We can use AI to do the work but not to get the skills that come from doing the work. And without the skills, all we’re left with is a generation of people who can use tools and manage processes they don’t understand. We’ll have efficiency without competence, output without expertise, results without understanding. And the moment the tools break or the moment they subtly, quietly start giving us the wrong answers…we’ll realize we gave away something we can’t get back.
And by then, it will be too late.
So What Do We Do?
I run an AI consulting business, so I use AI every single day in my work. Instead of rejecting the technology altogether, this article is about the importance of using it wisely.
Here is my framework for working with AI:
Use AI for:
- Tasks you’ve already mastered (augmentation)
- Repetitive work where you can verify the output
- Generating options you can then evaluate
- Producing alternative approaches to your thinking
Don’t use AI for anything where the journey matters more than the destination. This could be:
- Using AI to do your homework (you’ll never learn the topic)
- Writing articles from scratch (but use it to evaluate and critique your first draft)
- Generating any output you don’t have the sufficient knowledge to verify
Thanks for reading! This post is public so feel free to share it. And if you liked what you read, consider following me for more posts. Take care!