22/10/2025

I don't think people realize how vibecoding can't get anything meaningful in production.

You can't improve what you don't understand! The idea of staring at an LLM doing the work, not understanding what it did and just hoping it will work and not break anything else is just foolish.

When you're working on fixing issues, on taking something that works 95 times out of 100... to "always working", you need to understand the root causes, the effects of your fixes, the potential unintended consequences. Otherwise nothing guarantees you're not creating a mess that's worse than the initial problem.

That's also a risk for traditional engineering, but the difference is "understanding" what you're doing. Vibecoding has been sold as a way to build stuff even though you don't understand anything of what it does. "Everyone can be a builder! Skip the learning!"

That's BS. Everyone can build a pretty interface or luckily ship something that seems to be working on the surface. But you're being fooled if you think you can avoid engineering (someone involved who can understand the system and not break it further when doing changes)

25/05/2025

AI agents shall be used by people who can understand them.

The latest "vibe coding" trend lets us believe a non-technical person can be turned into an engineer.

For having tried, it's not the case. In the case of coding, AI agents misinterpret clear requirements, forget context, break what worked. The amount of supervision to get anything useful/functional/secure requires you to know what you're doing... so, being competent.

There's a big incentive (probably to raise money) to oversell AI agents capabilities.

As I've always said, AI assistants have clear limitations and while they can be of great help, the person using them must be able to practice judgement. And informed judgement based on professional knowledge.

So, the best use of AI is as assistants to people doing the job, rather than replacing the people doing the job.

This sounds less disruptive but it builds a more responsible use of AI, and as a bonus, people get to keep their jobs instead of losing them due to wrong expectations of what AI can do.

just saying!