Pythonium

Python, What else ?

Developer fatigue in the age of AI-generated code

It took me a while to get into coding with AI agents. I had no idea how much this would change my job as a developer. The way we develop software has changed considerably: generating a function, creating an interface, writing tests or even building an entire feature has become much faster.

When we're unsure which solution to choose, we can now implement all of them and then keep the one that makes the most sense. Something that simply wasn't possible before.

On paper, this is great news. Less time spent on repetitive tasks, less code to write manually, and more time available to focus on interesting problems.

But unfortunately, I've noticed, and apparently I'm not the only one, that a side effect has emerged: a certain weariness with our own profession, which no longer has much in common with what originally made us enjoy software development.

When coding becomes almost too easy

A few years ago, when I had to develop a feature, I would often spend time thinking about how to implement it, trying several approaches before getting a satisfactory result.

This research phase was part of the pleasure of software development. We could spend an hour on a problem, try several solutions, understand why something wasn't working and eventually find an elegant approach. And that was rewarding and enjoyable.

Today, a good part of this work can be delegated to AI. Even the most rewarding part. We describe what we want, challenge it a little, get a good-quality result, and move on to the next task. It's efficient, sometimes even impressive, but it also fundamentally changes our relationship with code.

We don't necessarily build things the same way anymore. We do more supervising, validate suggestions and fix what doesn't work.

And for some developers, myself included, this change can be quite unsettling.

From developer to code reviewer?

The most noticeable change I've experienced is the amount of time I spend reviewing code rather than writing it myself.

AI can generate dozens or even hundreds of lines of code across twenty or so files in just a few seconds. But that doesn't mean those lines are correct, suitable for the project or easy to maintain (for me, this is still a criterion, unlike what I read on LinkedIn).

We have to understand what has been generated, check the assumptions, identify potential side effects and make sure the result actually meets the requirements.

And this can become quite tiring over time. Reading and understanding code written by someone else already takes effort. So when you have to do this dozens of times a day, with code whose original design decisions you don't always fully understand, the cognitive load can become significant.

The satisfaction of finding a solution yourself

Another change is that finding the solution to a difficult problem yourself brings a certain sense of satisfaction. You've searched, experimented, sometimes got frustrated, but eventually you understand and solve the problem.

With AI, this satisfaction can sometimes disappear. When a solution arrives almost instantly, you go through fewer of these steps. You get the result, but you haven't necessarily experienced the process that allows you to fully understand it.

It's a bit like watching someone solve a puzzle for you. You know the answer, but you haven't really had the pleasure of finding it yourself.

Obviously, this doesn't mean we should go back to a time when we spent hours looking for a bug that AI might find in a few minutes. But I think it's a shame, and even rather depressing, to completely lose this intellectual dimension of the job, the very thing that made us want to do this work in the first place.

More productivity, more demands?

Productivity gains can quickly turn into new expectations from management. If an employee can now complete a task twice as fast thanks to AI, it's tempting to give them twice as much work. The time saved doesn't necessarily become free time. It can simply be used to increase the number of features expected.

We risk falling into a kind of productivity race where we have to produce more and more, simply because the tools make it possible. And that's not necessarily what developers are looking for.

Being able to finish a task faster should also give us time to think, test more thoroughly, document things or simply take a breather. Otherwise, what's the point of saving time?

The risk of gradually losing certain skills

What happens to our ability to develop software without assistance? By constantly asking AI to generate code, suggest architectures or solve problems, we may gradually lose certain instincts.

We may become less comfortable with documentation, less used to figuring things out on our own or less able to understand certain mechanisms in depth.

This isn't inevitable, and AI can also be a great learning tool. It allows us to discover new approaches, understand concepts and explore technologies we might not otherwise have taken the time to study.

But we need to remain vigilant. Systematically accepting an answer without trying to understand it is probably one of the worst habits we can develop. At least for now, I still make the effort. But for how much longer?

In conclusion

AI brings a lot to software development. I would be in a poor position to claim otherwise. It helps us save time, explore new possibilities and accomplish things that would have required much more effort in the past.

But we shouldn't reduce a developer's job to their ability to produce code alone. Searching, understanding, designing, experimenting and solving problems are also part of what makes this profession interesting (at least for me).

If tomorrow, our job essentially consists of making requests to an AI and then reviewing what it produces, some developers may no longer find what they used to enjoy about software development. I'm one of them.

We'll see how things evolve over the next few years...




Popular posts