I always know when we're in an interesting point of a change when people are tripping over language to describe things.
It seems to me that everybody has a slightly different expectation of what vibe coding is, and what to expect from it. Same with AI generally. On the one end you get people insisting its going to change the World, and on the other you get "vibe debt" references and, well, this: https://www.youtube.com/shorts/qWJUWinWWnQ
My experience is that all AI today is good at helping experts be more productive by giving them first draft artefacts that might need editing, and sometimes that editing is extensive, sometimes maybe nothing.
That's it.
They are not "doing the job", they're assisting the job. People who expect AI to "do the job", are setting themselves up for failure, whether its filing paperwork in courts [0], writing book reviews [1], or writing software [2].
Sure you can use midjourney to produce some artwork for you, but if an artist used it, they'd be able to make it even better and fix it up in all sorts of ways, so maybe you will still need to pay the artist, they will just take less time and maybe lower their unit price of work.
If you are an expert at software engineering, and you are using AI, your job is changing, but it's not going away. I think every mature conversation about current agentic coding practices acknowledge this. And I think people who are getting good at this will be more productive, and they will get more done, but they're not going to be getting their job done in 4 hours and taking the rest of the week off, because they're just going to try and add more value to get paid more, or even just to keep their jobs. That's the treadmill of free market capitalism.
The "vibe coding" stance tries to counter this, but doesn't acknowledge the limitations of the technology. I think it can be taken in one of two ways: ignore the code and be surprised at what comes out and have fun with the results, which I think is fine for a hobby, you do you; or, ignore the code because you don't need to understand it, and let's get rid of all the expensive developers, and hey, what could possibly go wrong in this production stack, which I think is going to end very badly for all involved.
While today the expectation of vibe coding with current technology is as ridiculous as me saying I'm going to replace my lawyer with a bot and YOLO it [3], I am not sure that will hold forever, or for the rest of my career, or for even the next year. The only way to find out is for experts to try it, and qualitatively and quantitatively assess the results. I'd love to see a vibe coding index which periodically tries to get apps of increasing complexity built and then assess the results, and we track this over time. At some point maybe we'll get through a tipping point, and we can have conversations about what that means.
I think we might need a better term for it though, that isn't AI-assisted development (as the intention is that its not assisting you, it's doing it). Natural language programming? Entirely Generative programming?
It seems to me that everybody has a slightly different expectation of what vibe coding is, and what to expect from it. Same with AI generally. On the one end you get people insisting its going to change the World, and on the other you get "vibe debt" references and, well, this: https://www.youtube.com/shorts/qWJUWinWWnQ
My experience is that all AI today is good at helping experts be more productive by giving them first draft artefacts that might need editing, and sometimes that editing is extensive, sometimes maybe nothing.
That's it.
They are not "doing the job", they're assisting the job. People who expect AI to "do the job", are setting themselves up for failure, whether its filing paperwork in courts [0], writing book reviews [1], or writing software [2].
Sure you can use midjourney to produce some artwork for you, but if an artist used it, they'd be able to make it even better and fix it up in all sorts of ways, so maybe you will still need to pay the artist, they will just take less time and maybe lower their unit price of work.
If you are an expert at software engineering, and you are using AI, your job is changing, but it's not going away. I think every mature conversation about current agentic coding practices acknowledge this. And I think people who are getting good at this will be more productive, and they will get more done, but they're not going to be getting their job done in 4 hours and taking the rest of the week off, because they're just going to try and add more value to get paid more, or even just to keep their jobs. That's the treadmill of free market capitalism.
The "vibe coding" stance tries to counter this, but doesn't acknowledge the limitations of the technology. I think it can be taken in one of two ways: ignore the code and be surprised at what comes out and have fun with the results, which I think is fine for a hobby, you do you; or, ignore the code because you don't need to understand it, and let's get rid of all the expensive developers, and hey, what could possibly go wrong in this production stack, which I think is going to end very badly for all involved.
While today the expectation of vibe coding with current technology is as ridiculous as me saying I'm going to replace my lawyer with a bot and YOLO it [3], I am not sure that will hold forever, or for the rest of my career, or for even the next year. The only way to find out is for experts to try it, and qualitatively and quantitatively assess the results. I'd love to see a vibe coding index which periodically tries to get apps of increasing complexity built and then assess the results, and we track this over time. At some point maybe we'll get through a tipping point, and we can have conversations about what that means.
I think we might need a better term for it though, that isn't AI-assisted development (as the intention is that its not assisting you, it's doing it). Natural language programming? Entirely Generative programming?
[0] https://www.independent.co.uk/news/lawyers-london-england-hi...
[1] https://chicago.suntimes.com/opinion/2025/05/29/lessons-apol...
[2] https://cyberplace.social/@GossiTheDog/114546174330023840
[3] https://petapixel.com/2025/04/14/man-tries-to-use-ai-generat...