AI has become, over the last few years, an impossible technology to ignore. Almost everyone ignored it at some point, and now most developers have adopted it for daily use—both professionally and personally—including myself.
However, AI is not meant for code editors. It should stay in the browser.
Coding requires deep thinking about every line while maintaining the overall structure and design of the system. The AI we have today cannot do this consistently or deeply (and no, this is not a prompt issue). When embedded directly into the editor, it mainly helps with small, repetitive tasks while overlooking the broader architecture of the project.
Yes, it can reduce typing time, and for less experienced developers it may even produce better-looking code than they could write themselves. But it can—and will—introduce subtle bugs. This creates a tradeoff between speed and code quality and security.
So should we still include AI in code editors because it saves time? No. And here’s why.
First, typing speed is almost never the bottleneck in software development. Second, the time saved by AI is often lost debugging subtle issues introduced by generated code and overlooked by the developer.
For less experienced/skilled developers (and sometimes even experienced/skilled ones, aka the x10 devs), AI can generate new ideas or better implementations. However, as a project grows, the developer may lose track of what is actually happening because they do not deeply understand the generated code. This leads to even more time spent asking the AI to explain the code, searching documentation, and filling gaps in knowledge about algorithms, libraries, or techniques they never truly learned.
While that learning process is not wasted—and can improve the developer’s skills—it does not require AI to be embedded in the editor. All of this can (and should) happen in the browser.
Another reason AI should not be in code editors is that it encourages laziness over time. Even experienced developers risk skill atrophy. Gradually, they may become similar to less experienced developers: relying on AI for things they once knew, then wasting time debugging AI-generated bugs they failed to notice.
In the future, AI may or may not become capable of reliably writing large, complex systems. If that happens, we may not even need code editors—only interfaces to observe and validate the output.
Until then: no AI in your code editor.
Note 2: This post was reviewed and rewritten by an AI—in the browser.
Note 3: Sorry ${PREFERRED_AI_EDITOR}
Top comments (9)
I totally agree with you!
👍
I using AI ( gemini, codex, copilot ) cli version, which is give much different AI helped development experience. I cannot able to compare the code editor version, which I not used.
Why?
Because my initial idea is the code development need to be independent from code editors. Many times I just use simple vim for editing ( I don't use too many vim motions just: o, O, a, A, i, I, y, p, dd, x, :, esc ), I am also familiar to direct codewriting in browser console. So my suggestion is for code development do not need to stack some dedicated code editor functionality without AI or with.
AI cli strong ability on larger project to just give a prompt and AI is work on a problem ( mainly refactoring legacy codebase ) on agent mode around 10min -> 1.5h and do a crazy amount codebase change. Just the PR time is fare more higher, but I think this is will worth.
Finally: terminal is your friend.
nice one, badr! thanks 💯
It's a difficult subject for sure. For myself, it's goes beyond the title of this post, as I am very privacy-focused with regard to the applications and services I use. I have no interest using Visual Studio Code, for example. Not only is it a Microsoft product but it is becoming more entrenched with Co-Pilot. It has telemetry too. Yes, much of this can be turned-off or ignored but that won't always be the case. I have considered VSCodium, though I'm also taking a firm look at Pulsar Editor as well. If I was to go for an AI supported editor, it would probably be with Zed.
Again, this is very much a personal standpoint and based on the desire to migrate from Big Tech. This isn't something everyone will want to do or be able to.
sometimes the suggestions deviate me from my thinking and that irritates a lot; it’s a gamble where sometimes it is very productive, and other times it is the reason a 2-minute task turns into 2 hours.
The key difference between AI in the editor and in the browser is that it can change your project files. And the differences end there (apart from the user interface)
The quality of the AI result depends on the prompt, so writing a complex program using AI is like writing your own book; the more detailed you describe it, the better the result. AI can be bad for those who are too lazy to write prompts, but good for those who clearly define their goals.
I am not sure I agree. AI can help create better code faster, both for inexperienced and experienced developers. At least from my experience and from what I learn from talking to my fellow experienced developer friends (aka old guys).
But, and here I see your points, it should be applied wisely if the project should be maintainable.
And it doesn't replace traditional quality controls such as static code analysis, unit test coverage, clean modular code, well-designed APIs, code reviews, penetration testing etc.
"I'm not convinced AI's the scurvy dog we think it is. It's a double-edged sword, capable of slicing through code with ease but also introducing subtle bugs that can send your ship to Davy Jones' locker. Can we trust AI to chart the course of our code, or should we keep it anchored in the browser?"
This comment was written by AI too lol (o′┏▽┓｀o)
I see your point, and it makes total sense to me... At least in some cases.
Like most things in software development, the answer depends on context.
Do I think it’s okay to use AI embedded in your editor to build a simple web page? Yes.
Do I think it’s okay to use it to analyze your code and suggest improvements with more context than a generic web search? Absolutely.
But do I think it’s okay to turn on auto-accept and let AI drive your development? Hard no.
It’s a powerful technology and can be extremely helpful when used wisely. If you don’t have the patience or self-control to use it that way, you’re probably better off not using it at all, your results will likely be better.