How AI changed the way we code

Computer programmers since always was being taught that we should minimize redundancy. From super popular books we were learning that code with duplication is bad.

And this approach is really good and very friendly to our brain, because people love abstraction, especially engineers. If you see the same pattern in few places, it’s very natural you instantly want to abstract it and reuse.

And it makes a perfect sense. This way you no longer need to think how you have styled CTA button on your page before you add another one. You just import the button component from your library and change only text and link.

But LLMs operates totally different way.

For LLM this abstracted button is just another file that it has to:

And if you think about it, it also makes sense. Just treat it as if the code you’re working with was write by a bad programmer, who weirdly named components, placed them all in a single directory. This way you’d also prefer if every piece of code was written in the same file, because then you’d be able to read everything at least without jumping through different files.

And that’s how LLM codes - it prefers self-sufficient files with all needed code defined in one place. That’s why you see now people making fun of vibecoded apps with overly packed files taking care of everything.

But is it really that bad?

Well, yes and no. It really depends on who you will be reading the code.

If you’re vibe coding your app and work alone on the code - it’s completely fine to allow such redundancy. It can actually speed up the process and lower the costs of tokens.

It’s also great for MVPs. In business it’s really important to act fast, publish unpolished projects and get to the market ASAP. So overthinking code quality and refactoring same class 5 times is not worth doing if you’re building MVP that aims to test if this app makes sense.

But if you’re building app that is running on production, or you’re working on it with someone else - code quality and readability matters. In that case, you should make sure that not only AI can understand your app, but you should be the one who understand it even more.

Following DRY (don’t repeat yourself) and KISS (keep it simple, stupid) principles that we learn from big software development books are really useful. That way maintaining, debugging, refactoring and developing app is much much easier.

And if you switch from vibe coding to AI assisted coding - you no longer rely on LLM finding that file you though of, you just reference specific part of codes from your knowledge.

One tip that can help you develop clean code while vibecoding is using good templates / boilerplates. Although they have some kind of learning curve for humans.

Nowadays developer can have hard times.

On the one hand, LLMs (especially combined with agentic tools) are the best chance to become famous 10x (or 100x) developer. We can now become ultra rich CEO of own startup with countless AI employees just waiting to code your every single idea with no hestitation and for a fraction of junior’s salary.

On the other hand, the same developers that are super productive with Claude Code or Cursor start to feel anxiety, because now even people who can’t code are able to get fully functional app from their idea within a single prompt.

But we’re definitely not there yet. It’s still not enough to write what you want to get in order to run a business around software. Developers are still needed. Even Anthropic, who claims that AI will take replace software engineers in 6-12 months, is hiring engineers now. However we can also see that the market is shrinking and tech giants are firing people.

So how to survive those times as a developer? And how take advantage of the IT industry and earn even more money?

AI changed coding forever. Developers switched from StackOverflow to asking AI. If you’re coding without AI assistant - you fall behind.

And as AI changed coding, it will change also other industries. Maybe not in such a drastic way, but it is affecting a lot of businesses even now.

The best way to stay on the surface is to learn AI and use it effectively. This way you will have head start against those to refuse using it.

But it’s not enough to learn and use it as you’ve learned it. AI is still in super early stage and everything is changing weekly. Even if you don’t want to test every single new model, it’s good to keep track of trends and learn from other people’ experiments.

Not only learn AI, but learn more WITH AI.

Using artificial intelligence is probably the fastest way to learn new things and skills. You no longer need to deep through the internet if you face a problem when learning. Even for free you can get instant answer adjusted to your needs, where you can also ask following questions to understand it better.

AI gives you also way to learn in practice the same way you would sit next to your mentor watching him at work.

Learning devops? No problem, just think of what you want to learn and ask Claude Code to do it. Sit, watch and learn. Ask questions if something is unclear. If that’s not enough, ask other LLMs what can be done differently and why.

With proper AI usage, you can learn faster than you ever thought you can.

In order to be competetive in today’s developers community you need to either:

There’s no other way. Everything between will be punished by the market.

If you’re a generalist, with AI you’re entire IT army. You can build frontend, backend, database, deploy app and even promote it.

If you want to be an expert, pick one niche. Niche it down even more. And become a real specialist in this, learn every piece of technology involved, meet every other specialist in this niche. Just be the guy that pops to everyone’s mind if they think about your niche.

Takeaway: Learn AI and learn with AI. Generalize or become ultra niche specialist

There’s no a single place where you can learn AI. What’s more, if you learned AI year ago and haven’t update your knowledge, it’s pretty outdated.

With such intensive pace of technology development, you can’t rely on any learning focused content. If someone will have enough time to master a thing, digest it enough so as to prepare well thought course for you - you’re already late.

The best way to learn in such conditions is to:

If there’s a new cool model - test it out.

Code a lot of apps, even if they’re stupid. That’s the way you learn in practice, by engaging yourself in and by making mistakes.

And AI is not a math, physics or grammar. There are no strict rules, it’s not fully deterministic. In order to be the expert you need to know what can you do, know the possible tools and have a little experience using it.