What's new

It has been a long time since I’ve written anything for this blog. The original idea was for this blog to shed some light on the stuff I had been working on, but I found out that maybe I did not want to show everything that I had been working on to everyone. Texting one or two friends was enough and also simpler than writing an entire article.

But here we are again.

My problem with AI

Since the advent of artificial intelligence for serious code work, I have been really struggling to learn something new.

And when I say “serious code work”, I do not mean simply copying and pasting your code into a chat and then waiting for the AI to perfectly solve your problem. When I talk about serious code work, I mean the boom in agentic coding between the end of 2025 and the start of 2026, when AIs started being capable of searching for context by themselves and running arbitrary commands on your machine… right there, something changed.

Before that, we already had some cases of people doing this (like Cursor). That let us code basic stuff / boilerplates with AI, but for more complex work it was still faster to program stuff by hand. The models could also search for more context by themselves before, but they had a very small context window, and this forced them to forget stuff really fast.

In the period between the end of 2025 and the start of 2026, most of these problems had been solved. Now the agents could run arbitrary commands on your machine and understand the problem by themselves, search for context better and forget less, and, of course, the models themselves had become better and more optimized for this type of work.

My previous paragraph makes it sound like everything became wonderful, but that is not the whole truth. “AI” just became good enough.

AI becomes good enough

When I say “good enough”, I mean that nowadays writing code by hand is 100% optional, and it is also considerably slower and more tiring than just asking an LLM to do the work instead.

Nowadays, at work, I only use AI. 100% of my code is made by AI, and most of the tests to see if the AI has made any mistakes are also made by an AI.

My job is, for the most part, just to act one level above the code, thinking about the problem that must be solved. A lot of the time, during this process, I also do this while talking with an AI.

The hardest “work” I do is reviewing that same code made by the AI, seeing if it fits the current project / architecture, if there are any mistakes… etc.

And when I read it, I see that the code is not excellent, but good enough, and probably if I had written the same code by hand, it would have taken a lot more time to do the same thing. In the end, the problem is solved and the client is happy anyway.

Easy life

Can I be honest? Doing all this stuff is easy… oh, it is so easy. I really feel like I haven’t worked hard since I started using Codex and Claude Code at the start of 2026.

Not everything is perfect. Since the start of this period, I have run into some issues that the AI could not solve, or cases where the end result would just suck. But in the end, the solution was always the same: prompt again, and eventually the issue would get solved.

It could take some time, but hard? No, it was not at all. You could say the hard part is enduring all the laziness and boredom that this brings.

And it is in this boredom and laziness that the issue appears.

Difficulties in improving

Like I said at the start of the article, my biggest issue is continuing to get better and learning new stuff.

I feel like, since the start of 2026, I haven’t learned anything new. I just stay here sitting still, waiting for the chef to make burgers and fries while I sit at the table waiting for the food.

That kills me slowly. How can I excel at my craft if the only thing I can do is ask someone to do the stuff for me? This guy is also always available and can do it faster than I ever could.

Should I, instead of learning how to cook, learn how to ask? Well, if I learn how to ask, I could, instead of asking for burgers and fries, ask for handmade ravioli with fresh pasta dough.

But asking is such a passive action. At the end of the day, everyone can just ask. It cannot be that simply doing something like this is enough to make us excel at something.

From chef to waiter

Now let’s assume that I’m a waiter. Maybe, for me to excel, I would need to understand my clients and what they want, allowing me to give much better and more efficient instructions to the chef. Maybe I could also start cooking sometimes, so that when I need to give instructions to the chef, I can be more precise, making the whole process faster and providing a higher-quality end result.

Well, without a doubt, that would make me a better waiter, but note one thing: waiter.

Did I really learn gastronomy just to become a simple waiter?

No. I started because I wanted to become a head chef.

Honestly? I know how to become an excellent waiter, but a head chef? Right now, I have no idea how I can become something like that…

So for now, I will just force myself to become an excellent waiter.

Conclusion

Putting all the abstract analogies aside, I’m trying to get back to improving at software engineering and really stop feeling stagnant, so this blog is a new attempt at doing that.

But sadly, I cannot learn by just reading stuff. I need to do stuff, discuss, create, develop.

And the discussion part will be on this blog. It will become almost like a notebook of stuff I have been studying.

The plan is basically to write articles to consolidate stuff in my memory better. Writing forces me to put into words things that maybe were stored in an abstract way in my head.

So that’s basically it. This blog, if everything goes right, will be filled with stuff I’m learning about.

See you later!