I don’t know what’s going on. I always wanted to solve problems and create things, but now that I have access to a tool that’s able to speed up the whole design and development process… I’m just bored.

I don’t feel the passion of building anymore, mainly because I’m not building at all. There’s no craft in letting something else build on your behalf, there’s no real thought or design in writing prompts. There’s no engineering or challenge left. It’s just: prompt, evaluate the output, adjust, and repeat. What the hell is this? Sometimes I use them to actually learn something - but it’s a different way of learning. It’s like reading a summary or a description of some math formula, without applying it - quite pointless.

I have always been a “maker” - in my own way. I used to study topics in depth, spending hours and sleepless nights thinking about problems and solutions. I’ve experienced that great feeling of “this is a great idea” coming out of nowhere while doing unrelated activities, and having to rush back at my desk to write it down and implement it. That was great - it was rewarding, it was really meaningful. My brain was working and I felt satisfied every time I looked at what I created. Perhaps this is the result of 5 years of university studying computer engineering, but I guess not, it’s something I’ve always done and the engineering mindset just fit.

Over the years, I’ve built software of all kinds: from ML libraries to web development tools, passing through websites, wrote an infinite number of scripts for Linux system administration, set up servers, created online communities of tech enthusiast, designed database architectures. And a lot more - my guilty pleasure of mixing languages and technologies that I didn’t know, just for the fun of experiment, learning, challenge myself, and spend some time having fun.

Now, all of that feels… gone? I can just have an idea, ask the model-of-the-month to implement it, and - wow - it’s there. Written in no time and with code quality that improves month after month. If I add guardrails, rules, skills, and design documents upfront, the model generates pretty much the same thing I would have written (not really - but you got the idea). Sure, it still depends on the size and complexity of the idea, but anyway, the possibility of prototyping in no time is nice — but terrible?

To be clear, no one is forcing me to do this on my free time. On the job - and apparently across corporate tech in general - there’s this huge pressure for using those tools to not “get left behind”. Left behind what, exactly? I’m quite sure that if someone finds a real productivity metric for the software development, that takes into account speed of execution, technical debt, maintainability, and cost - LLMs won’t shine, and their introduction and usage without brain ^{is just the generation of endless technical debt.}

Still, I find myself using LLMs locally because they are a tool, and as I did for any other tool and technology I discovered and used over the years, I just want to use them and master them. But this time the usage is really boring… Luckily in this process, at least, I had some fun setting up a machine for local inference.

What’s the point?

The point was never just going from A to B. The point has always been to have fun while going from A to B, learn something new along the way and gain savvy.

That’s the core issue: savvy. And I am losing it.

According to Merriam-Webster:

savvy: having or showing perception, comprehension, or shrewdness especially in practical matters.

Savvy is the fundamental idea for me - learning by doing. Do, make mistakes, learn from them, repeat, and grow.

With the LLMs every time there’s a mistake, I don’t learn. I can spot the mistake because I have some experience (lucky enough to have been born 34 years ago), I can pinpoint the mistake and let the LLM fix it.

The shitty thing here is that LLM providers can steal my savviness to train their models - so I’m not even going to see those mistakes anymore, and I can’t keep my savviness trained.

So… what now?

I don’t know. In the process of using LLMs I’m still learning new things and having some fun — at least I set up a Linux machine for local inference, and that part was a blast. But this blog post is mainly just a rant, humanly written, that’s going to be used by some company to train some LLM and in the future used by someone else to generate something similar, when prompted “Write me an article that targets the niche of software engineers that are bored by the AI usage. Do not make mistakes”.