Is AI making me dumber?

I've been a heavy user of Cursor since September 2024 (I'm still running on their old pricing model). This year I started to use Claude Code mainly, but for some tasks I rely on JetBrains Junie if I want to stay within RustRover IDE.

Yesterday I had a thought popping into my head: Is AI just making me dumber?

I was trying to debug an endless render loop in JavaScript that I couldn't wrap my, or rather GPT5's, head around. The AI kept suggesting the same fix, over and over. I'd tell it why that didn't work. It would apologize. Then suggest the same thing with slightly different wording. The context window grew to thousands of tokens. I started a fresh session - same loop. At the end of the session I was JUST SHOUTING IN ALL CAPS TO THE PROMPT WINDOW THAT IT SHOULD NOW FIX THAT GODDAMN BUG... I went mental.

I must have looked like one of those TikTok doom-scrolling zombies.

All that, for a bug that I could've resolved myself in (probably) 20 minutes of time.

Instead, I spent thousands of tokens, and 1.5h on this little bug. Which ended up in a one-line fix.

It felt like a complete waste of time, tokens and Co2.

As someone who started writing code for money at 14 years old, I know that I have the technical knowledge, the skills and the debugging capabilities myself. But these days I find myself going back and forth with an LLM for coding tasks a lot.

Feeling lost in my own codebase

I remember how back then I always knew which files to open, quickly jump around and find my way through the code. In case of a bug I could fairly easily pinpoint where it could've originated from. I knew how the system worked, in-and-out.

Now it's easier than ever to get lost in your own codebase, if you let AI just do its thing.

Back then, whenever I joined a new project I struggled a bit to orient myself around the existing codebase. The getting to know part, learning the ins and outs. That has changed. And probably for the better. It's easier to explore code bases and understand relations between functions and explore parts of a codebase, thanks to AI.

But if you use it to write 100% of the code, it becomes very easy to just "vibe-all-day". And eventually get lost. If you just "Tab" "Tab" "Accept" "Accept", without thinking, this can happen quick.

As Matthias said:

[...] If that happens, then you went too far, but sadly it's too late. Now the codebase feels kinda "dirty", like fast food.

And you even get the feeling that it's "ok" because the re-generating of that code is cheap and fast anyway.

Losing my skill

I honed the skill of coding for years. I started with some basic PHP and JavaScript. Over the years I learned many languages (PHP, JavaScript/TypeScript, Python, C, C++, Rust) and would consider myself an experienced software engineer.

When ChatGPT came out I started to use it now and then for some coding tasks, but it never felt like it was good enough. Things changed when Cursor dropped. The integration right into the IDE, and the model advancements were a noticeable improvement. I used it more and more throughout my days, created UI mockups with v0 and started to build AI powered workflows and apps for myself and clients.

Then Claude Code dropped, and I was stunned by the way it worked. The feeling of an actual fully autonomous agent. Their TODO list feature was a killer, and their way of searching the codebase was so different to what Cursor implemented.

But with Claude Code a feeling started to creep in: I'm losing my skill. It's like the side effect of a drug.

I wrote less and less code myself. Instead, I reviewed tons of code, learned how to write specification files and behaved like a product manager, not a software developer.

But it's the act of writing out the code, using the language's syntax, remembering constructs, and the idiosynchrises... that actually hone and practice the skill of... well... writing code.

I never wanted to be a manager. I don't like being a manager. And don't aim to be one. I want to be a developer. A maker.

Even as someone who has written a lot of code and would call myself a very experienced developer, I sometimes find myself lost in codebases. Forgetting, how to implement a certain feature, or how to do something which I have done in the past, but maybe not in the last couple of months.

That is a reason why I'm actively trying to reduce my usage of AI as a developer. I'm not cutting it down to zero. I'm still using AI, but I find it very important to control your usage, to take a step back from time to time, still write code on your own because you really lose your skills otherwise.

Maybe the daily practice that Matthias Endler is doing (writing Rust code in the Playground without even an LSP) is the way to go.

Challenge for you: Write out a simple program without any LSP / Autocomplete / AI support in your favorite programming language, that reads a file and prints the file contents line by line. Can you do it? It's pretty simple, but I bet most of the people will struggle.

If you choose Rust, play around in the Rust playground with this fiddle.

I tried this challenge myself. Luckily, I was still competent enough to do it. I still knew the syntax in-and-out. But even then, it took me a bit of trial and error to get it to compile. That was a wake-up call.

I'm still questioning if in five years this skill is completely obsolete or not, but for now I really want to keep it.

Losing my love for the craft

AI automation for your job sounds awesome. Until you realize you're automating the parts you actually love. - Ole Lehmann

If you've come so far you can already tell: I love coding.

I love to make myself a cup of coffee, sit down at the computer in the morning, put my noise-cancelling headphone on, turn the techno up, open the IDE and start coding.

It's a ritual. It's a daily practice. Or it was a daily practice. One that I lost for a while and am now trying to build back up.

I love to type on my keyboard (that's why I have a variety of super weird ergo keyboards). I love to build things from code.

Thinking through complex problems and solving them with code is satisfying. The coding is the fun part. But with AI the fun part gets lost. You type in a few words into a TUI (Terminal UI), with some back and forth and Claude does it for you.

Sometimes even a lot faster than you could. And if you parallelize the work with multiple sub-agents even faster.

Now you're sitting there, busting out code in rapid-fire mode, but in the end it feels like you haven't done a thing. The same goes for writing. It just feels like it's not your own work... And I don't feel proud of that work.

And here's the thing that really scares me: When the joy goes, when the craft becomes just "getting output," what am I even doing this for? The code was never just a means to an end for me. It was the end itself. Now I'm questioning what makes me useful at all.

Now I ask myself, what does it even mean to write code yourself? Is it necessary that you remember all the syntax of Rust in-and-out?

What am I even getting paid for? My critical, and system thinking capabilities? The bigger context window in my head? Sometimes I'm not sure.

I've asked other people for their point of view:

Oh yes, I've noticed that too. It's especially frustrating when you spend the whole day tinkering with a function and then get a 3,000-line PR from a colleague that you're supposed to review. Unfortunately, I don't know the answer to that. - Matthias Endler

Another colleague Lucca stated it differently:

If the AI can fully generate the code, then I don't want to have anything to do with the project anyways. I do not want to write code, that can be generated... Development is largely a symbiotic process between the developer and the codebase - and if that’s true, then real quality can only emerge from that relationship, meaning that neither AI nor outsourcing pose a real threat to important software engineering work. - Lucca Hellriegel

I have not yet found a real solution to this problem. I'm exploring whether or not building and shipping something valuable is the real deal and coding was always just a means of getting to that point.

The fear of job displacement

I've done some mentoring at MasterSchool for a few months and saw many students struggle to find an entry-level job. But even senior engineers these days struggle to find a job, and many are used to high-salaries and tons of benefits. The tech-hiring peak in 2020/2021 was insane. But these days? It's quite the opposite. You see headlines in the news of lay-offs in tech. And then you see this chart:

A chart showing the downfall of tech-hiring in the past few months.

There are many arguments of why this trend is not only related to the AI hype, and CEO's saying that 80% of code will be written by AI at the end of 2025, but also being related to the market situation. Anyway, it's a problem for many people out there. And people worry, just as I do.

I do worry about the future of software engineering as a craft. As many others have already compared it to: It's like a craftsman losing his craft. Like people that worked at assembly lines being replaced by machines and robots.

If we rely on AI and the foundation models from companies like OpenAI and Anthropic, what is the side-effect of that? Yes local models exist, and they become pretty good and more "affordable" to run on non-super-high-end computers, but it's not even close to existing providers like Anthropic or OpenAI in my opinion.

Falling behind

You open YouTube and see: "Idea to App in 52 Minutes", "How I built a SaaS that makes 15k $ MRR in a weekend using AI", "How this guy copied a billion-dollar company's software using AI and now makes 20k $ MRR". It's a perfect cocktail of FOMO, anxiety, bold clickbaity headlines and a lot of slop.

You open LinkedIn and see: "I got laid off", "I'm looking for a new job", "I'm on the lookout for a new job as a software engineer for 4 months now".

Combine this with the fear of job displacement and you get a perfect cocktail of anxiety. At least that's what it's like for me.

At the moment I'm working more than ever. Most days are filled with screen time, 11-12 hours a day according to the screen time app (and no that's not on the phone). Yet it still feels like I'm falling behind. The headlines for AI 10x'ing your productivity as an engineer sound appealing, and it makes you feel like you have to use it to stay in the game.

And honestly? I, too, think so. If you don't grow past what you've already known you're not going to make it.

If I had to decide whether I want to get into the field of computer science I'd still do it, but only because I actually enjoy it. When I was at university there were many people without a real interest in computer science, they were there because of the promised job security in the future. I'm not so sure if those people are going to make it.

I hope I'm going to make it haha. Back then everyone said: "Oh you're in IT! You have a safe job and a good income. Nice." But I worry that in five years of time it's not going to be that safe anymore. The market is saturated. Less junior roles. Less pay. Making it less attractive.

But what if no juniors follow? Is AI taking over the whole IT sector and nobody is going to work in it anymore? I don't think so, but maybe the whole sector will shrink and condense itself.

If we make ourselves dependent on AI, we're screwed. We'll be in the hands of big companies that rightfully so only have their profits in mind. So is the benefit of using AI only short-term? Matthias also wrote a nice blog post about this titled Paolo the plumber.

What matters to you?

I could end this on a high-note. But I won't. Instead, I'll be keeping the positives for a dedicated blog post coming in the next three weeks, after I completed 12 months of fully leveraging AI tools.

What matters to you? Maybe it's the fulfillment you get of doing it "the hard way", because that is what matters to you. Maybe it's going full-on-AI and finding your life's meaning somewhere else entirely. There is no right or wrong answer. There is only your answer.

What do I want to learn, even if AI gets better at it? Do I still want to have the ability of the skill itself, even though AI will be better at it?

Sometimes the answer will be yes, because of personal growth eg. that supersedes any ROI. Sometimes the answer will be no, and that's ok too.

For me, right now, that answer is this: Use AI less. Use it more targeted. Not zero AI - I'm not going back to punching punchcards. But I'm reclaiming the parts I actually love. I'm writing more code myself. I'm sitting with problems longer before reaching for the AI prompt window. I'm treating it like a reference, not a replacement.

Will this make me "less productive" by AI marketing standards? Probably. Will it make me a worse engineer? I don't think so. Will it make me feel like myself again? That's what I'm betting on.

I'd love to talk about this with other devs and even non-devs (if you're reading this as well). Hit me up on X or LinkedIn and we can chat.

Links: