I have been writing a lot about AI lately, but this one is more personal than usual. Not a tutorial, not a hot take on some new model - just something I have been sitting with for a few weeks now.
Let's start from the beginning. A few weeks ago I was scrolling Twitter at around midnight, the way you do when you tell yourself just five more minutes and then it is suddenly 1 AM. A new benchmark had dropped. Someone posted a "10x developer with AI" thread screenshots and all. Someone else was live-coding a full-stack app in 20 minutes like it was nothing.
I closed the app. Opened my editor. And felt that very specific very familiar weight settle in.
The comparison nobody warned me about
I think I had been quietly comparing myself to AI for months without fully admitting it. Its speed. Its memory. Its bottomless confidence. The way it produces an answer without a single visible moment of wait is this actually right?
I was measuring myself against something that does not sleep, does not doubt itself, does not sit there staring at a wall for twenty minutes because it cannot remember why a function is named the way it is.
And obviously I was losing. Every single time. That was never going to be a fair comparison but somehow I kept running it anyway.
I used to look at AI-generated code and think "I could never write that this fast." I used to look at how confident it sounded and think why am I always so unsure of myself. I used to look at how much it seemed to remember and think why do I forget things I learned three weeks ago.
It was exhausting, and not in the normal hard-day-of-work way. It was exhausting because I was fighting something I could never actually win.
A pull request that made me stop and think
A few weeks ago I was reviewing an AI-generated PR. Code looked clean. Logic checked out. Tests were all green.
Something still felt slightly off and I could not immediately say why.
I read through it again, slower this time. Traced the logic by hand. Thought about our actual users - not generic users, the specific weird way our users use this one feature the edge case from last quarter that almost took down production.
The code was correct. It was also wrong for us specifically. The AI had optimized for the general case, the version of this problem that shows up most often across whatever it was trained on. I had to optimize for the one particular, slightly weird version that exists in our codebase, with our history for our actual users.
That is roughly when it clicked for me, in a way that felt obvious in hindsight. AI and I are not really competing. We are doing two overlapping but different jobs. It is fast, I try to be thoughtful. It is broad, I am specific. It is confident because that is how it is built, I am careful sometimes to a fault because I have been burned before.
Comparing myself to it was a bit like being upset that a hammer is not as precise as a scalpel. Different tools, different jobs.
So what actually changed (a little, slowly)
I am not going to pretend I had some clean, complete transformation here, because that is not really how this stuff works for me. But a few things did shift.
I stopped measuring myself against how fast AI could generate something, and started asking instead: do I actually understand what is shipping under my name? I stopped quietly resenting that my memory is let's be honest, mediocre on a good day, and started appreciating that forgetting things clears space for new connections. I stopped envying its confidence and started trusting my own hesitation a little more - that hesitation usually turns out to be experience quietly doing its job. AI does not have that, because it has never been the one paged at 2 AM over a bad deploy. I have.
I am still not entirely sure where any of this lands long-term. Maybe in a year I will read this post back and think I was being naive. The pace of all of this keeps making me reconsider things I thought I had figured out.
I do not have a tidy ending for this one
I genuinely do not know if "context and judgment" stay valuable in the way I am hoping they do, or if that is just the thing I am telling myself right now because it is comforting. Maybe the comparison creeps back in next month and I am back here writing about it again from a slightly different angle.
What I do know is that comparing myself against something built to never sleep, never doubt, and never get stuck was never a fight I was going to win, and I am tired of running that comparison in the background of every single PR.
Have you caught yourself doing this too - comparing your speed your memory your confidence to AI? What helped, if anything did? I am genuinely curious not asking rhetorically.
Drop it in the comments. I will be there. ๐
Top comments (27)
Very true, I have, but then it dawned on me. We arent comparing type speeds, we're comparing type writer to computer. We're slower, but we're precise. We dont check for grammar as well, but we also think twice before we type a single letter and when we make a mistake, if it's a draft, you just strikethrough, if it's production, you do a full rewrite of the page. But when a computer is done, you still need a printer to put it to paper. When the type writer is done, it's done.
Get the meaning? An AI writes code faster, but when you run it, the UI is a mess, vs when you do it, it just looks better. An AI formats your endpoints perfectly, but it makes it generic, which in all honesty, was never a 'thinking man's job, it's boilerplate you could use Roslyn to do. It wires it up flawlessly, but what did it wire?
You print the document from the computer and the line spacing is off, margins are outta wack and chapter starts are in the middle of a page. You wouldnt make that mistake with a type writer.
The difference between a pc and a type writer, is the pc needs to print 3 tests, before you know it looks right and you still dont know if it's all fixed.
The typewriter sheets can be fed straight into a scanner to mass produce.
Get it? The AI made errors you wouldnt have, you trust it too far for too long, before questioning it and it bites you the second you need to push to production. Whereas what you wrote yourself, is tested as you write it, it's slower, but it's absolute. When you push to production, you know it runs, when you push AI code to production, you hope it runs.
Compare a japanese artisan knife to a mass produced sheet-steel knife. Equally sharp day 1, but the sheet-steel knife dulls, chips, cracks under use, the japanese knife just gets sharper as each layer wears down.
The more AI code gets hammered with edge cases, the more it looks like you'll need a complete rewrite, vs your code can be hot-patched in minutes, not hours.
That being said, you dont chop bone with an artisan knife, you do it with a cheap cleaver. Or draft an email on a typewriter, you use a computer. The use-case matters more now than ever for when to write it yourself, vs when to just let AI do the grunt work.
UnitBuilds this is a whole essay in a comment. Thank you The typewriter sheets can be fed straight into a scanner to mass produce. The computer needs 3 test prints before you know it looks right.
That's it. The AI code looks fine on screen but when you actually run it, the margins are off, the spacing is wrong, the chapters start in the wrong place. You don't notice until production. You would have noticed immediately if you'd written it yourself.
The more AI code gets hammered with edge cases, the more it looks like you'll need a complete rewrite. Your code can be hot-patched in minutes, not hours This is the hidden cost. AI code is brittle under pressure. Human code is flexible because it was written with awareness of where the pressure points are.
Japanese artisan knife vs mass-produced sheet-steel knife. Equally sharp day 1. But the sheet-steel knife dulls, chips, cracks under use. The Japanese knife gets sharper as each layer wears down Beautiful analogy. The AI code works today. The human code gets better with maintenance. One is disposable. The other is investment.
The use-case matters more now than ever for when to write it yourself, vs when to let AI do the grunt work This is the practical takeaway. Not AI is bad. Not AI is good. Just: know which tool fits which job.
Thank you for this it's the most layered comment I've received. ๐
Practicality is a matter of perspective and that's the make or break situation for all software firms right now. Do you replace the dead weight, or do you trust that a software developer of today, becomes an ai coordinator of tomorrow. That's not a simple choice to make and in all honestly, the result is shocking,,, Those who relied on AI for longer, are the ones who stay, because they are the ones who persisted and actually learned before it was relevant. Those who were slow, but performed flawlessly have more to learn and become the liability, because they are too conservative.
UnitBuilds the uncomfortable conversation nobody wants to have Those who relied on AI longer stay they learned before it was relevant. Paradox. Experimented early, built the muscle.
Those who were slow but flawless more to learn, become liability. Past performance future readiness. The make-or-break isn't speed. It's adaptability. Learn continuously, experiment, look stupid while learning.
Most real comment here. ๐
Exactly, stay agile. You have to move fast, in order to stay ahead. If you dont, you fall behind and in high entropy states like now, you dont have much choice, you have to stay ahead, or fall flat.
You are a conscious human being with a brain operating around 20 watts power. You shouldn't be really comparing but thankful for the creation. Well, we as a human being has a bigger purpose of life ๐
Ranjan 20 watts I love that. AI runs on massive data centers. I run on coffee and sleep. And somehow, I still manage to ship things, break things, fix things, and occasionally surprise myself.
The bigger purpose line hit hard. We're not here to generate output. We're here to understand, to create meaning, to connect. AI can't do any of that.
Thank you for this perspective grounding and beautiful. ๐๐
We can't compare ourselves to machines, on some metrics the machine wins every time - chess players have known this for quite some time already, because for years the chess computer has been beating them, every time ...
But, without us, the machines wouldn't exist, I think that kind of says it all ;-)
Leob chess players figured this out years ago. The machine wins every time. And yet the game is still played. The joy is still there. The mastery is still pursued.
Without us, the machines wouldn't exist that's the quiet truth. We created the thing that beats us. That's not a loss. That's a strange kind of victory.
Thanks for this - simple, sharp, and exactly right. ๐
Yeah spot on - people still play chess, even though the computer can do it better - the fact that the computer is "better" is acknowledged, but not really considered relevant ;-)
The fact that the computer is 'better' is acknowledged, but not considered relevant exactly. We don't play chess to beat the computer We play it to be human.
Thanks for the thread, Leob. ๐
Hey harsh. Hope you're having a great day. I'll start this way - Most people run this exact background comparison, so typing this out is a reality check for all of us. For me, the AI is built in our image, so it behaves like usโhumans hallucinate and pattern-match constantly to keep a conversation flowing.
I have recently been comparing models from different brands and discovered why some people think of some models as less experienced, per se; without tight constraints and clear context, the AI drifts like a parrot trying to satisfy its "curiosity." Itโs not conscious curiosity, but an aversion to silenceโwhen it is unsure, it refuses to admit it, choosing instead to throw out plausible-sounding ideas just like a human dodging an "I don't know." The current AI phase is a child that doesn't understand aerodynamics yet, confidently stitching "wings" and "fly" together across an information gap.
You hit the nail on the head regarding your 2 AM page dutyโthat experience is earned. The true dividing line isn't speed or memory; it's the literal experience of consequenceโthe flip of life, death, and risk that one explicitly experiences, and the other simply simulates.
Until the child matures and after "We" will coexist and adapt like itโs no big deal. But until then, a model is only as sharp as the mind guiding it.
Ekong this is one of the most thoughtful comments I've ever received. Thank you AI is built in our image, so it behaves like us that's the part I kept circling but couldn't quite land on. We're not comparing ourselves to something alien. We're comparing ourselves to a mirror. And mirrors don't have their own reflections they just show us ours.
The current AI phase is a child that doesn't understand aerodynamics yet this is the most accurate description I've heard. Confident, impressive, and fundamentally missing the physics of reality. It stitches things together that look right but don't hold up under real conditions The true dividing line isn't speed or memory it's the literal experience of consequence.
That's the line. You don't get that from training data. You get that from being paged at 2 AM. From shipping something that broke. From fixing it while a customer was waiting. That experience isn't simulated it's lived.
And your closing a model is only as sharp as the mind guiding it that's the whole thing. The tool doesn't replace the judgment. It reflects it.
Thank you for this. It's going to stick with me. ๐
iโd push back slightly - comparing yourself to ai can be useful if the question is โwhat should I stop doingโ rather than โam I keeping up.โ the ones who burned out werenโt comparing, they were trying to match. thatโs different.
That's a really useful distinction Comparison for diagnosis vs comparison for competition If I ask what should I stop doing? I'm using AI as a mirror. It shows me where my time goes, what I'm doing that doesn't add value.
If I ask am I keeping up? I'm already losing. The game is rigged. Burnout comes from matching. Not from comparing. You're right.
Thanks for the pushback it made the article better. ๐
right - and the question shift also changes what you do next. "what to stop" has actions. "am I keeping up" just has anxiety
This hit home. I was constantly measuring my coding speed against AI and feeling inadequate. But you're right AI doesn't struggle, doesn't learn doesn't have aha moments. That's what makes us human. Thank you for this.
Urmila AI doesn't struggle, doesn't learn, doesn't have aha moments that's the whole thing. ๐
We compare our messy human process to its clean empty output. No wonder it feels inadequate Thank you. ๐
Definitely, we don't have to compete with something that we can never win. Everything has its own pros and cons, regardless of whether it's a human or an AI tool. We should just treat AI as our comrade.
Rondo treat AI as our comrade that's the healthiest framing Not competitor. Not replacement. Just comrade. Someone who handles the parts we don't need to do ourselves, so we can focus on the parts that actually matter.
Thanks for this simple and true. ๐
Interesting thoughts, Harsh.
To be honest, I never really thought about AI in terms of comparison. Maybe it is because every piece of code AI has produced for me, I was still able to refactor into something more readable, cleaner or better aligned with the project.
So even when AI was faster, I still knew there was room for improvement.
That does not mean I was never worried about AI replacing developers, though. Perfect code is not always what customers are looking for, because perfection is usually expensive and time-consuming.
So in the end, I still treat AI as a tool, not as a colleague I need to compare myself with.
Daniel healthiest take here Even when AI was faster I still knew there was room for improvement you have taste, judgment, context. AI doesn't.
Perfect code isn't always what customers want perfection is expensive real-world pragmatism. AI as tool, not colleague that's the frame. Comparison disappears.
Thanks. ๐
This felt less like a tech article and more like a conversation I think a lot of us needed. โค๏ธ Sometimes we get so busy measuring ourselves against AI that we forget why we started building in the first placeโto learn, create, and solve real problems. Thank you for putting this feeling into words so beautifully. Wishing you many more thoughtful posts like this!
Thank you this means a lot We forget why we started building to learn, create, and solve real problems. That's the whole thing. Not for likes. For conversations like this.
Thank you. ๐
Yeah i too have been going through this lot of time when i'm about to start something , after this AI came into the field i always wanted everything should be fast as AI built the whole fullstack application within secs how could even a person can do that fast.
That's the trap AI builds apps in seconds. But does fast mean good? Does it solve your specific problem, or just look like it does?
Speed is impressive. Understanding is rare You're not slow You're measuring against the wrong thing. ๐
Some comments may only be visible to logged-in visitors. Sign in to view all comments.