A Signal That Sent a Chill Down My Spine
Let me start with a line from the retrospective of someone who uses AI heavily every day. Reading it gave me a bit of a chill:
When I have a really enjoyable conversation with AI, it probably means I didn’t grow that day.
They explained it clearly: a pleasant conversation usually means no friction was encountered. Real growth is always accompanied by some kind of discomfort — forcing a vague idea into a clear sentence, working through a problem you can’t figure out, being jolted into rethinking by an objection. None of that feels “good.” They kept using AI to remove friction, and the more they used it, the smoother and more pleasant the conversations got — until they stopped and realized: friction was exactly what they needed most, and they’d deleted it with their own hands.
This line stings precisely because it exposes something we rarely admit: the biggest temptation of the AI era isn’t “getting dumber” — it’s “getting too comfortable.”
AI Is a Machine for Removing Friction
Think about every tagline in AI product marketing — faster, less effort, one-click generation, no learning curve required. Its entire value proposition is removing friction: turning what used to take real effort into something effortless, turning what used to get stuck into something smooth.
In a lot of situations, that’s purely good. Repetitive, trivial, mindless labor — formatting, translating, looking things up, compressing ten documents into a single page — that friction never had value to begin with. Removing it is liberation.
But here’s the problem: AI doesn’t distinguish between “friction that should be removed” and “friction that shouldn’t be.” It flattens both equally. Ask it to help format something, and it does. Ask it to help you “figure out what to do about this,” and it also does — but that discomfort of “not being able to figure it out” is exactly the growth zone you were supposed to walk through yourself. It sifts out the gold along with the sand.
Growth Comes Almost Entirely From Friction
Why can’t you just remove all friction? Because if you honestly recall every real leap forward you’ve ever made, almost all of them happened in uncomfortable moments.
Writing is the clearest example. Someone who spent over a decade in sales and was restarting mid-career was asked to write a long essay every week. At first, they had no idea what to write or how, and painfully forced themselves to record and reflect every day. Three months later, they could easily write a thousand or two thousand words, sitting quietly at their computer, entering a state of flow. They condensed that change into a formula: reflection + pain = progress.
Note that “pain” isn’t a figure of speech. Writing enables growth precisely because it forces you to turn the ideas in your head that you assume are clear into a sentence that actually holds up — and that process instantly exposes every vague spot, every logical leap, every unexamined assumption. That discomfort of “being forced to think clearly” is your mind building muscle.
Learning works the same way. You never really learn something in the comfort of “I heard it, I nodded, I got it.” You learn it in the awkwardness of “doing it yourself, getting stuck, debugging, getting it right.” Friction is the raw material of growth, not an obstacle to it.
So when AI makes all of this painless — you don’t have to write, don’t have to think, don’t have to get stuck — it also takes away your opportunity to build that muscle. What you’re outsourcing isn’t the labor. It’s growth itself.
The Danger Isn’t Using AI — It’s Outsourcing What Shouldn’t Be Outsourced
I want to be clear: this isn’t an essay telling you to use AI less. Friction that should be removed, should be removed boldly — saving effort in places with no growth content is exactly how you free up effort for the places that do have growth content.
The real danger is failing to tell the boundary apart, and outsourcing friction that should have been kept along with everything else. The outcome is very specific: you end up with an increasingly intelligent AI, and an increasingly hollow version of yourself. AI’s capability keeps rising while your judgment, expression, and thinking keep shrinking — because it’s been a long time since you last walked through the discomfort of “not being able to figure it out” yourself.
There’s a saying I strongly agree with: present yourself manually first, then use AI to discover your own boundaries. The order can’t be reversed. You have to run through it yourself first, hit that friction yourself first, arrive at “your version of the answer” first — only then is AI actually helping you accelerate. Otherwise, it’s just walking the path for you, while you stay exactly where you were. Friction is the only proof that “this time, I grew — AI didn’t grow for me.”
Telling Two Kinds of Friction Apart
So the question isn’t “should there be friction” — it’s telling apart which friction should be removed and which should be kept. Here’s a rough but useful line:
Friction that should be boldly removed (no growth content, pure consumption): formatting, information retrieval, translation, deduplication, compressing long material into a summary, initial classification — hand these to AI, the more aggressively the better.
Friction that should be actively kept (this is where growth lives): forcing a vague idea into clear text (writing), forming your own judgment on a question (thinking), deciding direction and trade-offs (decision-making), being jolted by an objection (self-examination). Even if AI could do these for you, you should do them yourself first.
A simple self-check: if doing this makes “you” stronger afterward, the friction should be kept; if doing it only completes “the task” while you remain unchanged, the friction should be removed.
Deliberately Inviting Friction Back In
Once you recognize this, “using AI” should be designed in reverse — not thinking about how to make everything smoother, but deliberately keeping friction at the critical points. A few concrete practices:
- Manual first, AI second. For any important thinking, write a rough version yourself first, then let AI polish it. Don’t open with “what do you think?”
- Write something every day. Writing is the highest-return way to actively manufacture friction — it forces you to think an idea all the way through, once a day.
- Give AI tasks, not directions. Rambling endlessly with AI about “how should I approach this” feels great, but that’s using pleasure to mask a lack of output. Treat it as an executor and hand it concrete tasks; keep “figuring out the direction” — the hard part — for yourself. (I wrote a separate essay on this: Give AI Tasks, Not Directions .)
- Let someone — or another AI — jolt you. Actively seek out objections and nitpicking reviewers. Comfortable agreement makes you stay put; uncomfortable pushback moves you forward.
A New Discipline: Treat Comfort as a Signal
Old-school discipline meant forcing yourself to do hard things. AI-era discipline adds another layer — actively choosing not to make certain things easy, at the exact moment everything else has become easy.
This requires a new kind of self-awareness: treat “comfort” as a signal worth being wary of. When you notice something has become effortless because of AI, when a conversation with it feels particularly smooth and pleasant — pause and ask yourself: am I making progress, or am I just comfortable?
Friction isn’t an enemy to overcome — it’s the nutrient of growth. AI can, and should, smooth away a huge amount of pointless resistance in your life; but leave yourself a zone that’s “deliberately hard to walk through,” where you’re still the one who thinks, expresses, and decides for yourself.
Because in the end, there’s only one standard for measuring all of this — not how much you got done using AI, but whether, in the process, you genuinely became a different person.
Related reading: Give AI Tasks, Not Directions | Series overview: From Information to Creation .
Responses