Why AI-Generated Code Will Hurt Both Customers and Companies
We’re nearing what feels like the top of the AI bubble. Over the last few years, the IT industry has been under constant fear that AI will take our jobs. Nowadays it seems like the narative has shifted a bit towards “AI is here to stay, but will be used as a tool instead of replacing jobs”.
There is still a lot of momentum going on though, companies are still pushing engineers to start using AI more and more in the hope that it will make them more productive. And to some degree it works and CEOs are happy that they can pay $100-$500/mo in AI credits to double or triple the output of a senior engineer instead of hiring more and paying full salaries. Not to mention that no one is hiring juniors to train them anymore.
While I do get the appeal and for some time I’ve also been hopeful that AI will take care of code generation instead of me having to manually type it using a keyboard, at this point I’m not holding my breath. It’s 2026 and we have the latest and greatest LLM models (Claude Opus 4.5, Google Gemini 3, OpenAI GPT 5-Codex), but they are still terrible at writing code. Every time a new model drops I get excited and try to use it for work and every time it fails spectacularly. If you give it a complex task to do it’s just plain bad. And even with smaller, very specific tasks it still fails 20% of the time. People who praise how good AI is at generating code, were not good at coding themselves in the first place. So at this point I don’t think it will get much better than this, the plateau has been real for a while now. Will it improve? Yes, but marginally, not enough to replace an actual senior developer.
Unfortunately we live in a world where companies are chasing short term profits instead of long term stability. And with this push towards using AI Agents for coding, things will only get worse before they get better. Most software products will show a decrease in quality and I think the best example of this is Microsoft who are taking pride in the fact that 30% of all new code shipped in Windows is AI generated … and it shows. A couple of months ago they shipped an update that bricked the shutdown functionality of the OS. The end users will pay the price of having to use worse products while spending the same (or even more) money for it. This will stop once users stop using said products and companies realise they need to focus on quality. We’ve seen this already with companies that replaced customer support hotlines with AI chatbots, people just hate them because most of the time they don’t work.
Is AI here to stay? For sure, but it will be just one other tool in our daily lives. Personally I couldn’t live anymore without the autocomplete from Copilot in VS Code. But other than that I wouldn’t trust it to write the code for me. Not to mention that there are already studies showing that using AI long term affects our critical thinking, so if we only let it write code, who is even going to be capable of reviewing the code in 10-20 years? Ask yourself a simple question. Would you fly in a plane if you knew the flight control software was written entirely using AI? I wouldn’t.
There’s also a hidden trap that I think most people don’t even realise yet. Ok, let’s say you are using LLMs to generate most of your code and you just review it. You are happy, your manager is happy, things get shipped faster, company profits increase and stonks go BRRRR. The thing is you’re not shipping code of the same quality as you did pre-AI. Why? Because when you just review the code you’re not capable of understanding it fully and I think anyone who has done PR reviews for a while will agree. I experienced this myself, I was reviewing AI generated code that looked really good, but then tend to skim through it without deep understanding of the problem. I just cannot do that unless I write the entire thing myself, to have it fully mapped in my brain. And this will make even great engineers worse at shipping good quality products.
For anyone who’s still wondering if a career in software engineering still makes sense. Yes, but the next few years will be rough. New grads will face one of the toughest job markets ever because companies don’t want to hire them anymore, hoping that AI will deliver on its promise and also the economy as whole is still not in a great shape. This will change in a few years when the AI bubble will pop, the dust will settle and companies will face a labour shortage again, because everyone switched to other careers and existing senior devs will start retiring.
Disclaimer: this blog post was written by a human
