AI Is Making You a Worse Coder—Here’s Why I Killed It in My Editor

My Evolving Take on AI-Powered Coding

In late 2022, I tried out AI coding tools for the first time—well before ChatGPT’s initial release. Seeing how effortlessly these models could explain obscure compiler errors and pinpoint issues in my C++ code felt like I’d discovered magic. By 2023, I was incorporating AI tools into my daily workflow, particularly GitHub Copilot and a handful of LLM extensions right inside my editor.

But then, by late 2024, I removed all AI integrations from my code editors. I still believe in AI’s potential to help developers, and I do occasionally use it. The question is: why don’t I keep AI embedded in my editor anymore?

How AI Tools Boosted (and Then Undermined) My Skills

When I first started using AI-powered code editors, it was like enabling Full Self-Driving on a Tesla—I barely had to do any work, just offer a little direction. Most of my tasks weren’t extremely complex, and AI’s suggestions felt like magic.

Then one day, on a personal account without Copilot or fancy AI plugins, I realized I’d grown oddly dependent on these tools. I hesitated after writing a function declaration, expecting the AI to fill in the rest. Even simple tasks like writing unit tests felt harder. Worse, I noticed that once I got used to AI suggestions, my intuition for coding decisions—even the smaller ones—began to slip. Sometimes the AI would get stuck, and I’d realized I wasn’t as quick at coming up with solutions on my own.

A big part of building real “senior” expertise, in my view, is honing that gut feeling—knowing which pointer type is best, when to rely on library features, or how to sense that something’s “off.” Using AI too much meant I wasn’t practicing those lower-level skills as often. And frankly, it started to show.

When I think about how some folks advocate “vibe coding”—just letting the AI handle everything—I can’t help but wonder how they expect to develop the instincts needed to handle the more complex issues that inevitably arise. As powerful as AI tools can be, sometimes you need real human insight, especially in those moments when production suddenly acts up, or certain frameworks behave unexpectedly.

Yes, You Can Thrive Without AI

It can feel like you’re falling behind if you don’t embrace AI. We’re bombarded with claims that “in 3–6 months” everyone who isn’t using AI will be left in the dust. But we’ve heard that line for years. My experience with Tesla’s “Full Self-Driving” package, paid for and still not fully delivered, has made me wary of hype and promises.

Right now, trusting AI to handle the bulk of your code on large, complex, or legacy projects usually isn’t realistic. AI can get stuck when dealing with custom internal DSLs, huge enterprise frameworks, or specialized game engines like Unreal. And in some industries, AI is off-limits for regulatory reasons.

Security-critical code in particular shouldn’t be handed over to an AI. Tacking on “and make it secure” to a prompt doesn’t magically protect you from vulnerabilities—some of which may have been baked into the training data. If security matters, you need to understand exactly what you’re doing. If one AI writes the code, another AI auto-approves it, and a third AI deploys it, that’s a recipe for disaster. Humans need to stay in the loop.

How I Use AI Now

I haven’t sworn off AI entirely. I just keep it separate from my editor so I have to manually feed in context. That little bit of friction means I’m still the one doing most of the work.

Sometimes I’ll ask an AI to convert a set of Go tests that use structs into a test map, or to rewrite something in SIMD. But I make sure I’m the one applying the changes in my codebase. It’s not enough to just approve the diff—I need to integrate it myself to really understand it. This approach also keeps costs low, since I’m not making a ton of unnecessary API calls.

I’ll also use AI to learn about unfamiliar or highly specific topics—like clarifying a snippet of assembly code, understanding what a shader is doing, or figuring out which books explain how to handle client/server desyncs in a custom game engine. The AI doesn’t always nail it, but it gives me better starting points than a typical search engine. And when I publish my own work (like on a personal website), I want it to be entirely human-created. AI-generated writing and images feel flat to me, and I prefer genuine human expression.

It’s About What You Love, Too

Yes, AI can save time, but speed and efficiency aren’t everything. Some of us actually enjoy the process of coding. When Deep Blue beat Garry Kasparov back in 1997, people didn’t stop playing chess. There’s joy in solving problems ourselves. That’s part of why I still write my own code.

To Anyone Learning to Code

If you rely entirely on AI, you risk becoming a perpetual junior developer—someone who never actually gains deeper skills. Keep pushing to learn the language you’re working in, the frameworks that power your apps, and the layers underneath them. It’s worth it, and frankly, it’s more fun to really know what you’re doing.

Even though AI might be “smarter,” you can’t blindly trust its output. Try going a few days without it now and then to keep your chops up. As your projects get more complex, AI’s suggestions often become less helpful. If you’re serious about a career in coding, skip the “vibe coding” approach. Positions that can be done by mindlessly prompting AI will be the first to go if AI truly takes over.

After all, if you can’t code without AI, can you really code?

Wrapping Up

Using AI inevitably trades knowledge for speed. That trade can be worthwhile in certain moments—but remember: even top athletes still do basic drills to stay in shape. The same is true in software development. Keep practicing the fundamentals so that you’re ready when bigger challenges arise.

Despite all the hype, AI still has limitations. We’re told it’ll replace everything in “3–6 months,” yet here we are. Companies have an incentive to stoke that fear of missing out, whether for investment or product sales. AI is just a tool—it’s what we do with it that matters. By not becoming completely dependent on AI, by making sure we understand every line of code that goes into production, and by staying curious, we stay in control of our work and our growth as developers.