I wanted to revisit a post I’d written about vibe coding making me feel dumb and the agent I built that could work with me at various levels to try and stem that atrophy. I think I ended that article by saying we’ll find out how it goes using this agent.
Truth is, I’ve barely used it. And honestly that’s because I’m totally drawn into the convenience of using Claude Code as my junior programmer, I suppose. As a hobbyist developer (now), getting something from my head into code quickly is the most important thing. I’m justifying the fact that vibe coding helps me do this. Whether this makes me dumb or not, I guess that’s a problem for future-me?
Losing control
Another thing I’d written about in that article was this realisation that all this vibe-coded stuff felt like it was getting away from me. What I mean is I was losing control. I was leaning on Claude Code and trusting it to produce stuff for me which, when I looked back at it, felt somewhat bloated and harder to reason about, let alone maintain.
Now for me, perhaps that’s OK. My projects aren’t hugely complex, needing to scale or be production-ready. I’d still like to understand how it works though and still enjoy the feeling of learning, and that’s why—at least some of the reasons why—I created the atrophy-fighting agent.
It’s one thing when it’s my own little projects. It’s another at enterprise scale.
The enterprise scale problem
Having worked over three decades in the industry, alarm bells start ringing when I hear things like Google using AI to generate at least 25% of their code. I assume the likes of Amazon and Microsoft are doing similar things. Can you imagine the amount of bloated code that’s being produced? Can you imagine how thoroughly unmaintainable that code might be?
I have vivid memories of watching development team leadership figure out how much time they should be devoting to technical debt. On the one hand, you’ve got product management pushing for features. On the other, you’ve got engineering pushing back just so they can fix all the stuff that was left behind—the bugs, refactoring code to make things more maintainable, architectures that ended up not working and needed to be redesigned. Technical debt was a huge thing as I remember it.
Oh by the way, design debt was big too. Our product design team was always pushing back to try and carve days into the plan to fix UX flows and patterns in our product because shortcuts were always being taken when our designs were implemented.
And now we’re adding… what? A 4× surge in code-cloning? AI that just copies and pastes similar blocks instead of creating elegant reusable logic? I’ve seen claims that it would take 61 billion work days to pay off the world’s current technical debt. That’s before we factor in the slop layer we’re building right now.
The age of disposable software
So there I am watching Nate Jones’s podcast where he opens with “the age of disposable software is here”. He goes on about the unconsidered consequences of this, and he does qualify that it depends on who’s doing it and what it’s for.
But this brought me back to this worry I have that we’re gonna need a lot of really well-versed senior developers who can literally babysit these AI workers churning out tons of bloated crap. But then you’ve got the situation where these corporations with huge development teams are dumping developers and relying more on generative AI to do their bidding. Less cost and no arguments.
How long is it going to take before we burn out senior developers because they’re being forced to tidy up the slop that AI produces?
Here’s the thing that really bothers me: some reports suggest senior engineers are around 19% slower when using AI tools. Not faster. Slower. They’re spending an average of 11 hours a week just correcting hallucinations—code that looks syntactically correct but contains logical landmines. I can’t imagine the stress for these folk.
The junior death spiral
Wouldn’t that time be better spent training junior developers to replace them when they do get burnt out? I mean seriously, what happens to the pipeline that takes new software engineers through the journey?
Entry-level hiring looks to have fallen sharply between 2023 and 2025. Companies apparently think AI can handle junior-level tasks. And sure, maybe it can. But if you don’t hire entry-level folk, you’ll end up with no senior developers within a few years — maybe less if you’re laying them off or forcing them to clean up AI slop. By the way, the boilerplate code that AI can create is the stuff junior developers do to get to know the frameworks right? Are we seriously thinking junior employees jump straight into complex architecture?!
I can’t help feeling that these large development organisations have cornered themselves and are going to have to rethink not only how they hire skilled practitioners but what the pipeline should be. Maybe generative AI will evolve to write such beautiful and elegant code that developers won’t be needed anymore. I’m sceptical. I know I’m biased.
There has always been masses of technical debt, but it must be growing exponentially… which makes developers extremely valuable to these companies right?
What I’m worried about
Look, don’t get me wrong. I do like AI. At least I’m very interested in what’s been done with it. I just wish the large corporations would think further ahead than just satisfying the immediate needs of the shareholders.
Because right now, what we’re getting is:
- Estimates like 45% of AI-generated code containing security vulnerabilities
- AI-generated pull requests with nearly double the issues of human-written code
- A massive backlog of technical debt that will eventually come due
- A generation of developers who never learned the fundamentals
- Senior developers being burnt out babysitting AI
Anyway, this was just something on my mind. I’m out of that industry now but still empathise a lot with it. I’m still vibe coding away on my little hobby projects, trying to remember to use that atrophy-fighting agent more often, and worrying about what all this means for the folks still in the trenches.
We’ll see how it all plays out, I guess.