"I'm going to call it Yaffle." That was the final line of my June New Atlas article, Domesticating AI: It's not coming, it's already here. It was a throwaway remark after spending some time with Home Assistant's increasingly capable AI voice assistant, Nabu.
I’m ecstatic to report that Yaffle is now real and answers to "Hey Yaffle." The problem was that Yaffle first needed to learn its own name. I'd tried this before, and failed, and the result was simple: I stopped using it. "OK Nabu" never felt right.
Teaching an AI assistant a custom wake word sounds like the sort of project you'd expect to find in the hands of a machine learning engineer. TensorFlow, speech datasets, model training – not exactly the sort of thing you'd tackle over a quiet weekend. Which presented a problem. I'm not a software engineer. Nor, I suspect, will one be turning up at my door to offer me a job. I started with Microsoft qualifications in the 1990s and have spent most of my career coaxing technology into doing useful things. I was happy enough with Notepad and HTML back in the day, but writing modern software? Training AI models? I'm just not a coder. My brain simply doesn't work that way. It put me well outside my comfort zone.
Or at least it did!
I'd been hearing the term "vibe coding," the latest catch-all phrase for building software by describing what you want an AI to do rather than writing every line yourself. The phrase has become almost as divisive as the technology behind it.
Spend half an hour in Reddit's programming communities and you'll understand why. It's easy to see why many programmers feel aggrieved. As a musician, I watched exactly the same thing happen to my own industry through the 1990s and 2000s, when increasingly capable software turned bedroom musicians into chart acts.
To many experienced developers, "vibe coding" has become shorthand for “people producing code they don't understand.” There are frequent calls for AI-generated code to be disclosed, warnings about security and maintainability, and no shortage of frustration that someone with little formal programming knowledge can suddenly produce something useful.
It would be easy to dismiss that as gatekeeping, but I don't think it is. If you've spent years mastering software engineering, algorithms, architecture, debugging and security, watching AI lower the barrier to entry must feel much as it did for professional musicians in the 1990s. Their concern is understandable, and experience still matters enormously. I just couldn't help feeling I'd seen this story before.
When university spat me out in the 1990s with an electro-acoustic music degree, producing professional recordings demanded training, specialist studios and dare I say, talent. Then Digital Audio Workstations like Cubase happened and suddenly, the bedroom became the recording studio.
Professionals were rightly aggrieved. Most of the music being produced was awful.
But some of it wasn't.
The same pattern repeated itself in photography. Skilled Photoshop artists spent years mastering masking, selections and retouching techniques, only to watch AI remove a complex background with a single click. Writers have seen grammar checkers become writing assistants. Graphic designers watched templates become design platforms. Video? That’s being transformed too.
Every generation of technology moves us one step further away from machinery and one step closer to intent.
Machine code became programming languages. Programming languages became frameworks. Frameworks became drag-and-drop builders. Now we're beginning to describe the destination instead of every step of the journey. We are rapidly approaching a world of near-instant, disposable applications built for the purpose of the moment, with the distance between inspiration and result rapidly decreasing. And it has all happened rather faster than anyone expected.
That's exactly what happened with Yaffle. I didn't sit down and ask AI to build me a wake word. Instead, I found myself acting as project manager and lead architect for a team that didn't really exist. I'd explain the problem, we'd discuss approaches, I'd test the result, discover what I'd broken, feed that back, and we'd try again. Somewhere along the way there were Python scripts, TensorFlow models, ESPHome firmware and more terminal windows than I'd care to admit. But that wasn't the conversation I was having. The conversation was always about the problem. How do I teach Yaffle to recognize its own name?
It let me have a conversation at the level of the problem I was trying to solve, translating that intent into code behind the scenes.
For the price of a monthly AI subscription, I suddenly had what felt like an impossibly talented senior programmer sitting beside me. Not someone who always got things right, but someone who could explain unfamiliar concepts, suggest approaches, write code, review mistakes and patiently iterate until we reached something that worked.
Yeah, patiently. It never rolled its eyes once.
Looking back, this journey didn't really begin with Yaffle. One small Home Assistant tweak led to another until I somehow found myself in VSCodium, the open-source version of Microsoft's code editor, where I discovered what AI-assisted development was really capable of. Very quickly, I was building things that once would have seemed absurdly beyond me. If you're wondering where all this starts, it isn't as intimidating as you might think. Modern editors such as VSCodium can host AI assistants. The workflow is simple: describe the problem, let the AI suggest a solution, test it, refine it, rinse and repeat until ... Ta-da! None of those projects started with the ambition of becoming a programmer. They started because I had a problem worth solving.
Perhaps that's why "vibe coding" feels so unsettling.
As a Brit, I am predictably hopeless at languages. Writing software always meant learning the computer's language. Increasingly, AI allows me to just use my own.
That doesn't make experienced programmers obsolete any more than Cubase eliminated musicians or Photoshop replaced photographers. If anything, it moves expertise up the value chain. The professionals still know what good looks like, still understand the trade-offs, and still recognize when the AI has produced garbage. Even Linus Torvalds, creator and lead maintainer of Linux, weighed into the argument this week. AI, he said, is "clearly a useful" tool; the challenge is making sure it helps maintainers rather than simply creating more work for them. Coming from someone responsible for one of the most consequential software projects on Earth, that feels like a fairly pragmatic assessment, not just hype.
Personally, I'm still confident I can tell the difference between a skilled musician, someone who simply knows Cubase inside out, and an AI-generated track, but the barrier to entry has undeniably shifted. I never woke up one morning thinking, "I'd like to become a software developer." I just wanted a better wake word than "OK Nabu." Before that I wanted a better Home Assistant dashboard. Before that I wanted a smarter lighting automation. Before that I wanted ... Every project simply led to the next one because the next problem suddenly looked solvable. That's the remarkable thing about vibe coding. It isn't that AI writes software. It's that, for the first time, millions of people can create software without first spending years learning the language of the machine.
Perhaps that's why it feels like cheating. AI didn't make me think like a programmer. It just lets me think like myself.
Or perhaps, like every great leap in abstraction before it, it's simply another moment where technology changes who gets to create.
