Skip to main content

Is AI making software engineering boring & us lazier?

Updated 5 min readMarkdown
Part of seriesJourneys
Is AI making software engineering boring & us lazier?

Before AI—or I could say before the big boom of Claude’s Opus 4.5—programming was pretty much the same as it had always been. People wrote their own code, collaborated on PR reviews, chatted, etc. People actually knew things in their own codebase—since they’d handwritten all that code, all them references and logic were injected into their brain, their mind.

Similar to writing something by hand in your notes, you remember that piece of information almost forever.

But it seems very different nowadays.

What's happening?

Now we just prompt and let AI write all them code. Some people actually review the code and make a few adjustments so other people can understand and maintain it. Some people don’t and just hit “commit” immediately.

This isn’t software engineering. SWE should be more than prompt & commit. You’re not a FDE (Forward Deployed Engineer). You got hired because you’re a Senior with experience—you can do the work and do it well. If everybody just prompts and commits, isn’t that a pretty solid way to show that you’re easy to replace? Even in the past, companies always said everyone was replaceable, and now that’s going to be simpler since they have access to AI.

Some people are willing to become a meat-proxy, or even go further: they just let AI reply to every PR comment, review, etc. I don’t know man, if I spend time reviewing PRs, that means I really care about the work, the product, and the company in general. But if I spend my time and get an AI-generated reply back, I’m not happy about that, and I don’t want to review that person’s PRs anymore.

When you work at a company, they hire you for your experience, knowledge, coding skills, etc., not just because you can use Claude and prompt—everybody can do that. You should think critically, be ready to step in, and be proactive at every stage of software development.

Finally, AI-generated code tends to come with super long comments that make it hard for non-native speakers like me to understand (and honestly, if I see AI-generated comments, I don’t want to read them anyway). But then the function and variable names it generates are the opposite: short and hard to understand when you read them. This isn’t sustainable or resilient. This is pure chaos.

What could we do better?

Review your generated code before committing

I don’t mind AI-generated code, as long as you own it, review it, and make the adjustments/changes to align with the conventions before opening a PR.

Respect the people reading your code so they can understand the what, the why, etc.

At the end of the day, it’s you or them who will have to jump into the code if problems/bugs/issues arise in production.

Write your own descriptions & reply to comments

Please, don’t be lazy on this part.

We are humans working together to develop our company’s software.

I don’t want to chat with your AI agents. Never.

We build software for humans, so the humans must know the software they are developing. Only then can greatness arrive.

Less AI hype

The more you hype AI, the less I know which part is scarier: AI or how little you actually deliver.

Hyping AI while delivering little or nothing feels like a joke.

Stay humble & stay foolish

Just remember to always improve yourself, learn more, code more, and gain more knowledge. AI or not, companies hire you for your knowledge & experience, not because you can use AI.

As for myself

I could say I’m proud that I keep learning/improving every day. AI is my companion/pair programmer to talk to and discuss tech, features, etc.

Company work

Backend code—I write it myself. I don’t want to be lazy. I want to understand what I wrote. I want to own it & take responsibility for it, etc.

Frontend code—AI does it, and I review the code and fully test it twice before committing to make sure there’s no sloppiness or bad code.

Personal projects

The stakes for these projects aren’t high, so I just delegate everything to AI and do the end-to-end testing, review schemas, etc. This is acceptable because it’s your own project, your decision.

But when it comes to company projects, make better decisions.

Advice for Interns/Freshers/Juniors

Try to stay away from asking AI to generate code. It’s super addictive and can lock you in your comfort zone, blocking you from learning more and gaining knowledge, etc.

For the first five years of your career, write your own code, no matter how long it takes. Learn and try out everything.

Ending

I really miss the past, when:

  • We had countless tech debates to compare and learn more about X and Y (e.g. language comparisons, microservices, etc.).
  • Show-and-tell sessions and sharing from colleagues were more thoughtful and full of knowledge, rather than model this, model that, new AI tool this, new AI tool that, etc.
  • We have more humans on the software engineering side. We learn, and we grow together.

Probably these aren't coming back, ever. Even so, I have a clear path that I want to go for myself.

17 views

About the author

  • develops awesome software, contributes to OSS, writes tech tips, and loves Vietnamese milk coffee!

The discrepancy you point out is really evident: produced comments that never end, followed by variable names that are so brief that it is impossible to understand their purpose. A model's first draft does not bother me, but I change the names and remarks before committing so I can still figure things out later.

KKi Taits6h ago

The mismatch you mention is so real: generated comments that run on forever, then variable names so short you can't tell what they do. I don't mind a first draft from a model, but I rewrite the names and comments before I commit so I can still find my way around later.

That’s great work & thoughtful of you. The more we think about the people who’ll deal with our code later, the more effort we put into making it readable. Time is not a strict requirement nowadays, but super high quality is.

Let’s keep that up 💪

The review rule is the part I will keep. If I cannot explain what the generated code does, I do not really own it, and I will be the one debugging it later. I also liked the split you use: write the backend yourself to stay sharp, and let AI help on the frontend only after a real review and testing. That keeps the skill and the speed in the right places.

Thanks!

In this AI age, getting too comfortable & lazy will get us nowhere. Staying sharp, keeping up the practice, and learning will always take us somewhere good. I don’t think it’s much different from the pre-AI days, but the pressure is definitely higher.

More from this blog