LLMs have come a long way since the launch of ChatGPT.

At first, you had to tell a model exactly what to do. Now, they understand much more, and they’re good at producing code.

We see them everywhere: website builders, LinkedIn post generators, tools promising you can create digital products in one click, no coding required.

Suddenly, people who never built a system can pilot a machine to do what they imagine.

But there are two big problems.

First, the machine is no engineer. It can suggest but it can’t decide. Feed it the wrong context, and it leads you into traps: bad stack choices, technical debt, poor security, high costs.

Second, debugging. I’ve worked with LLMs for three years. They suck at fixing bugs. Literally.

Maybe they lack knowledge of the codebase. Maybe it’s because they are prediction machines, not logic machines.

Chaos, not code

Either way, they’ll spit out 500 lines of code in a second. Bugs, unwanted features, hallucinated functions, missing libraries: the messy work lands on you.

Double the machine output, and double - maybe quadruple - the human cleanup. The work only skilled professionals can do.

LLMs also need quality data to improve. Data they can’t produce themselves. Every time you correct them, every time you surprise them with a solution, you teach them. You enrich them. Skills only humans can provide.

Some influencers say intellectual skills don’t matter in the age of AI.

Wrong.

They matter more than ever.

For sure, the industry will be harder for newcomers. First jobs in tech will be harder because code-for-code tasks, once given to juniors, are now done by machines.

If you want to stay relevant, you need to level up:

• Level up your ability to build and understand systems.
• Level up your productivity.
• Level up your ability to repair the mess machines leave behind.

Owning your code, owning your systems

I started with two skills: English, to sharpen my thinking and reach others - hence this blog - and coding from scratch, back to basics: good-old HTML and JavaScript, no frameworks, fewer layers.

Simple, clean, understandable code - so I own it, not the machine.

I level up to stay in control. Because once you stop understanding what the machine is doing, you lose control.

And losing control is the starting point of obsolescence - as an organization, as a professional, as a human being.

We need simpler systems. Minimum tech, maximum clarity. We must understand what we do and what the machine does.

The simpler the system, the cheaper, the easier to understand, the longer it lasts.

Building smarter machines only matters if it makes us wiser. Level up, simplify, and lead the process - before the machine runs the show.