Is vibe coding the new gateway to technical debt?

If programming were nothing more than taking an idea and turning it into running code, AI-assisted development might fully deliver on its promise. But AI’s ability to transform natural language prompts into runnable software ultimately highlights the importance of non-mechanical aspects of programming.

Rather than reducing or eliminating the human element in software development, AI reveals the human traits that are key to successful projects. One of these is the ability to sense when a piece of software is becoming unmanageable, and to mitigate the damage.

Such discernment differentiates novice and veteran developers. Increasingly, it also differentiates software written by humans from AI-generated code.

What vibe coding is not

By now, most developers know what vibe coding is. Many of us are using it in some part of our everyday workflow. AI-assistance being a part of the software development life cycle went from revolutionary to passé in about a week. It is extraordinary how quickly software developers can absorb shocks to the industry and just keep going.

But some wags keep missing this key factor: Developers are the ones adopting and applying the technology. We can’t be replaced because at the end of the day, we are the ones holding the reins that guide the tools.

The future is AI-empowered development, not AI-replaced developers. The age-old skills that make a good developer valuable become even more important when writing and pushing functioning code gets cheaper. Leave out the developer, and you just get mountains of technical debt.

You think requirements are easy?

The big idea in AI-driven development is that now we can just build applications by describing them in plain English. The funny thing is, describing what an application does is one of the hardest parts of software development; it’s called requirements gathering.

Anyone who has spent time defining requirements knows it’s a real swampland. It’s full of “I’ll know it when I see it,” which really means, “I’ll know that’s not it when I see it.” Bridging between the technical programming and end-user fit is notoriously hard to do. Good developers are the ones who can walk between those two worlds.

But now we are riding a vibe. A vibe, in this case, is an unwritten requirement. It is always changing—and with AI, we can keep manifesting these whims at a good clip. But while we are projecting our intentions into code that we don’t see, we are producing hidden effects that add up to masses of technical debt. Eventually, it will all come back to bite us.

What developers actually do

AI researcher Gary Marcus recently posted an article on Substack, Is vibe coding dying?, where he referenced a comment from a disillusioned AI coder:

I just want to say that I am giving up on creating anything anymore. I was trying to create my little project, but every time there are more and more errors and I am sick of it. I am working on it for about 3 months, I do not have any experience with coding and was doing everything through AI (Cursor, ChatGPT etc.). But everytime I want to change a liiiiitle thing, I kill 4 days debugging other things that go south.

So I do not have any more energy in me to work on this. It is hopeless. AI is still just soooooo stupid and it will fix one thing but destroy 10 other things in your code. I am really sad, because I was enjoying it in the beginnings but now it is just pain and rage. Hat down for those people, who can create something and it is working without coding knowledge.

The reason this quote is so perfect will be clear to most software developers. This is a non-coder who believes their project failed due to their inexperience and use of AI. But developers know the “I changed X, and way over there in Y and X something broke” is a classic pitfall of software development.

Software is a bizarrely interrelated complex of things, like a quantumly entangled system. Managing this madness is a big part of what successful software developers get paid to do.

Vibe coding and technical debt

Don’t get me wrong: I appreciate the unbridled creativity that can be found in just riding a vibe on an AI platform. If you haven’t already tried it, I suggest sitting down with Roo Code and seeing how it feels to blast out working software with just a few keystrokes.

At first, vibe coding is intoxicating. You can rapidly produce all the basic application infrastructure without even thinking about it. It’s like driving a 4×4 that sails over speed bumps. Pretty soon, though, you will find yourself off-roading in the ravines of wack-a-mole fix-and-break, like the above user did. Suddenly, that landscape of magically functional code becomes a yawning chasm of technical debt that you have to figure out. And if you don’t have the coding background to understand what you’re up against, you will drown in it.

Sure, you can try using AI to fix the things that are breaking, but have you tried it? Have you ever been stuck with an AI assistant confidently running you and your code around in circles? Even with something like Gemini CLI and DevTools integration (where the AI has access to the server and client-side outputs) it can so easily descend into a maddening cycle. In the end, you are mocked by your own unwillingness to roll up your sleeves and do some work.

It’s certainly one of the strangest experiences I’ve had with a computer: relying on my own hard-won instincts to ferret out root problems the AI itself obscures.

Be careful how you use it

Some might be tempted to accuse me of being anti-AI, which is not true. I love AI for coding. I’d even say (and have said) it brings back some of the joy and sense of possibility of the early days of the Internet. By dealing with a lot of the formality of coding, AI brings more ambitious ideas and efforts into scope. It lets developers spend more time in the zone of creativity.

If I had to choose one thing that is the most compelling about AI-coding, it would be the ability to quickly scale from nothing. The moment when I get a whole, functioning something based on not much more than an idea I described? That’s a real thrill.

Weirdly, AI also makes me feel less alone at times; like there is another voice in the room.

If in the end that is what we mean by “vibe coding,” I’m all for it. I like to be in the flow just as much as anyone else. The key is to be in the flow without unwittingly amassing junkyards of bad code. At the end of the day, AI will magnify whatever we put into it. It’s like trading a handsaw for a chainsaw: Better be careful how you use it.

The balancing act hasn’t changed

Programming has always been part engineering and part craft. A good software developer brings together these two parts in one mind. Modern AI helps with both the creative and mechanical aspects of software development, but a human is still needed to unite them with understanding.

The balancing act of software development hasn’t changed; it is just operating at a higher order with AI.