I already kind of feel the mental decline from everyday use of AI.

The Muscle Memory Problem

“I had some issues where I forgot how to implement a Laravel API and it scared the shit out of me.”

I can feel it happening. The muscle memory I spent years building, the instinctive recall of syntax, the automatic pattern recognition, the ability to read a codebase and feel where the bug lives, it’s eroding. The more I lean on AI to write code, the more I have to consciously read and re-read what’s in front of me. And here’s the cruel irony: the more I read AI-generated code, the less efficient I get at reading it. Mental fatigue compounds. I predicted this decline months ago, but predicting something and living it are two very different things.

I’m not alone in this. Developers across the industry are sounding the same alarm. One engineer on Reddit put it bluntly: “I had some issues where I forgot how to implement a Laravel API and it scared the shit out of me. I’ve been a software engineer for many years and it feels like I am back before I ever wrote a single line of code.” Another described it as “mentally outsourcing thinking in general.” Research is starting to back this up too, studies are finding that critical thinking skills genuinely degrade with heavy AI reliance. This isn’t paranoia. It’s atrophy.

The Silent Bugs Nobody Catches

“if the human reviewing the code is losing the ability to spot these issues, the guardrails don’t matter.”

And the AI itself? It’s not as good as the hype suggests. I was reading an article on Medium where a developer described how an AI-driven refactor quietly removed a null safety check. No fanfare, no warning, just a subtle deletion that slipped through review. This is happening every day. Data from code analysis tools confirms that error handling gaps are nearly twice as common in AI-generated pull requests compared to human-written ones. Missing null checks, ignored boundary conditions, swallowed exceptions, the AI optimizes for code that looks correct, not code that is correct.

Sure, we can improve our prompts, add better guardrails, and train the AI to be more careful. But that doesn’t solve the deeper problem: if the human reviewing the code is losing the ability to spot these issues, the guardrails don’t matter. You can’t catch what you can’t see.

We Lost the Joy, Not Just the Skill

“Now the job is increasingly about reviewing generated code, a task that is cognitively demanding but emotionally unrewarding.”

Here’s what nobody talks about enough. People who became software engineers tend to be deeply analytical. We’re the kind of people who got a dopamine hit from untangling a gnarly bug at 2 AM, from writing a clever algorithm and watching it work, from the sheer satisfaction of figuring something out. That was the reward loop. That was the thing that made the hard parts worth it.

AI has quietly dismantled that loop. Now the job is increasingly about reviewing generated code, a task that is cognitively demanding but emotionally unrewarding. You’re not solving problems anymore; you’re auditing someone else’s homework. And that “someone” is a probabilistic model that doesn’t understand your business logic, your architecture, or why that null check was there in the first place. The creative authority has shifted, and with it, the joy.

The Workplace Makes It Worse

“You’re caught between the expectation of speed and the reality of risk.”

Conversely, the average software engineer at a company doesn’t even get the consolation prize I get when I use AI on personal projects. When I build something I love, a tool I actually want to use, a product I personally relate to, there’s still joy in shipping fast with AI assistance. The product itself becomes the reward.

But at work? The pressure from leadership is relentless. Everyone assumes you should be 3x faster now because AI exists. Meanwhile, you can’t use AI the same way you do on personal projects. Enterprise codebases demand more scrutiny. Your team doesn’t share the same level of confidence in AI-generated code. There’s no consensus on how much to trust it, how much to verify, or who’s responsible when it breaks. You’re caught between the expectation of speed and the reality of risk, and the dopamine that used to come from writing smart, complex code and seeing it work? Gone. Replaced by anxiety about whether the AI quietly introduced a regression you’ll be blamed for.

Write Code by Hand Again

“A new language forces you to think from scratch, no muscle memory to fall back on, no autopilot.”

So here’s my practical take: while we’re in this transition phase, and make no mistake, it is a transition, not a destination, it makes sense to deliberately carve out time to write code by hand. Not because you’re anti-AI. I’m not. I’ve built multiple applications with AI assistance, shipped them fast, and I’m genuinely impressed by what these tools can do. But I also recognize that if I stop exercising the analytical muscles that made me a good engineer in the first place, I won’t be able to effectively use AI either. You can’t be a good code reviewer if you’ve forgotten how to be a good code writer.

I’ve been thinking about picking up Rust for exactly this reason. A new language forces you to think from scratch, no muscle memory to fall back on, no autopilot. It’s the programming equivalent of training with heavier weights. One developer I came across described learning C by hand and eventually being able to reproduce working code on paper without compiler warnings, something they were certain wouldn’t have happened if they’d leaned on LLMs during the learning phase. That resonates.

The Middle Ground Is the Point

“Are you using AI to amplify your intelligence, or are you using it to avoid thinking?”

I want to be clear: this isn’t a Luddite manifesto. AI coding tools are an unavoidable revolution, and refusing to use them entirely would be career suicide. But the developers who will thrive long-term are the ones who use AI to expand their capabilities without letting it replace their thinking. The ones who still understand their own codebases. The ones who can architect a system, not just prompt one into existence.

The question every engineer needs to ask themselves right now is simple: are you using AI to amplify your intelligence, or are you using it to avoid thinking? If you’re honest with yourself and the answer makes you uncomfortable, maybe it’s time to open a blank file, close the copilot, and write some code the hard way. Your brain will thank you.