What do you think about “vibecoding”? Is this a new way of building software, or another name for no-code development?

I have found vibecoding amazing for quick prototyping or getting a head start on a new project. But, when it comes to a critical production app, I am not ready to try this out. Every latest model I tested kept tinkering with parts of the code I never asked it to touch which is a recipe for nasty bugs. I am sure this will keep improving, especially once we pair it with rock-solid test suites that cover every crucial and edge case. For now, that cautious optimism comes from my own trial runs across several of my production project repos. Hope this helps.

Fundamentally, be aware of the "first spike" rule:

- You ask AI to generate stuff, and it does it, very fast, and mostly a-ok. That is the spike in productivity;
- Consequently, you ask AI to modify something of what it generated... and all hell breaks loose. That is the downside;
- Humans have to step in, and you face a need to have higher quality staff to fix the mess AI created. That is the plato.

Overall, using vibe-coding means you need less people - but more senior ones - to make it actually work.

Otherwise, what most businesses do is deploy their traditionally mid-grade-heavy teams who bring everything to the spike - then struggle after.

It's a neat idea.  It's fun to play around and see what the models come up with, but what they spit out at this point, sometimes doesn't even really run.  Not ready for production workloads for sure.  

Having looked at a lot of the code developed through Vibecoding I would say it's a fun hobby, and a nice experiment but it's not ready for any kind of production workloads. The reason is that most of the examples the code has been trained on is either sample code, that has zero security and minimal error handling built in, or it's from open sources projects most of which are incomplete or half-baked. 
The code produced at the moment with "vibecoding" is nice to get a few simple things working, and might be good for examples of how a particular language feature works, but you can't safely use it in a production setting. 

In almost all ways that count, vibe coding is merely the next, incremental step in the evolution of the practice of software development. Which means it is not a panacea: it simply is a tool that has ramifications when you use it. Some of these are good, some of these are bad and many of these ramifications are simply facts of life and are only good or bad in a specific context.

When I first started in IT, just after the invention of electricity, every programmer developed every single program from scratch. The only re-use we had was our own personal re-useable generic routines that we had developed (for example, to search a string for a pattern, or to perform range checking on an input field).

Then the web developed and, while most of us still wrote most of our own code, there were programmer's sites where we could post questions to each other about how to code for something or debug something. Sometimes that advice was great and sometimes it was not so great. And there were commercial libraries for things like UI, data handling and integration that took most of the coding out of those features.

Then, there were whole code repository sites available that a programmer could just search for to download whole routines for almost any language. Sometimes those routines were great and sometimes they were not so great. But, coupled with the commercial libraries and integrated development frameworks that would autogenerate the framework of the UI and core skeleton code, most programmers only custom-coded a minority of the total codebase - sometimes only a small minority - that constitutes the core business logic.

No we are getting AI to generate that core business logic, and sometimes the routines are great and sometimes they are not so great. Now, as always, the programmer is ultimately "Accountable" for the fitness of their software, even if they are not "Responsible" for all of it.

The failings of AI-generated software which are defects of the AI itself will soon vanish and only those defects related to the age-old issue of "garbage in - garbage out" will remain. While the ability to at least read code will be required of programmers for some time to come, the core skills of being a programmer in the near future are going to be related to their grasp of formal logic, being able to articulate boolean logic, pattern expression and similar intellectual skills so that they can accurately and unambiguously convey the specifications of the application to the AI. In other words, they will essentially provide robust and disciplined pseudo-code to the AI.