I sit somewhere in the middle of the AI hype cycle. Not a hater nor a fanatic.
I’ve been trying to make sense of AI by documenting my experiences and reacting to others.
Here are 4 issues I’ve found with using AI too much:
#1. It makes us lazier than usual. It’s tempting to go straight to a chat for a quick, unreliable answer. That’s the real danger.
#2. It steals the joy of figuring out problems. No more aha moments or happy dances. Just like asking someone to chew our food to be more productive.
#3. It makes you feel you’re cheating. Copy-pasting without understanding has always been cheating, even in the days of forums and StackOverflow.
AI generated code you don't understand is worse than technical debt. It's cognitive debt.
#4. It doesn’t let you build mental models to solve problems. It’s like asking someone else to go to the gym for us, then wondering why our muscles are still weak.
To build future-proof coding skills, check out Street-Smart Coding, the roadmap with 30 lessons I wish I had on my coding journey.
Top comments (7)
My thoughts are aligned perfectly with you on this topic. 🥰
While I just published a mostly AI generated project (link below 👇️) as part of the #hacktoberfest challenge, I really didn't get the joy of building.
The only joy I found was in the few parts that I did myself.
That said, this was a kind of a project I mostly want to hand over to something like an AI agent, so that I can focus on other important things. So for these sort of projects, I guess using AI is fine, for me at least.
Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass Submission 🌿
Birds.Quest - My Effortless Bird Watching Auto Blogging Buddy that Inspires Me to Go Back to Nature
Glad to hear/read that. No more aha moments or happy dances with AI...
The concept of “cognitive debt” is particularly compelling because it identifies a cost that traditional software metrics rarely capture: the gradual loss of understanding behind the code we produce.
I appreciated Baltasar's comparison to learning mathematics before relying on calculators, and Mohan's rule about only merging code he can explain. Both point toward an important distinction between using AI to extend our capabilities and using it to bypass the process through which those capabilities develop.
There's also a cognitive science dimension worth considering. Learning often depends on actively retrieving information, making predictions, encountering errors, and correcting our mental models. Research on desirable difficulties suggests that certain forms of effort, although slower initially, can strengthen long-term retention and understanding.
The challenge is that AI can remove precisely the friction that sometimes makes learning effective. But not all friction is valuable, either. Automating repetitive boilerplate may free up attention for deeper architectural decisions, while outsourcing every debugging challenge may gradually weaken independent problem-solving skills.
Perhaps the most useful question isn't “Did AI write this code?” but “Am I becoming a more capable developer through this process, or merely producing more code?”
Productivity measures what we accomplish today. Learning determines what we'll be capable of accomplishing tomorrow.
Great points!
Sure. Another problem is for students to cheat using AI. I tell them that, once they know enough about a programming language, or algorithms, they can use AI if they like to. But now when they know nothing.
Just like using a calculator in math classes...
"Cognitive debt" is the perfect name for it. As a solo founder I lean on AI a lot, and the rule I've landed on is that I only merge code I could explain to someone else. If I can't, I rewrite it by hand or ask the AI to walk me through it first. It's slower in the moment but has saved me from bugs I wouldn't have understood later.
Some comments may only be visible to logged-in visitors. Sign in to view all comments.