December 2024 — January 2025
It was December 23rd, 2024 — two days before Christmas — and I was supposed to be winding down. Instead, I was generating 10 websites in parallel.
Not because anyone asked me to. Not because I had a deadline. But because I'd stumbled into something that made my brain light up the way it hadn't since I first learned to code at 13 on my dad's PC back in 1993.
The post I wrote that night was almost casual: "Some people talk about development costs getting low with AI — I haven't fully decided yet, but I'm sure things are changing." Then I listed 10 website mockups, all first-response from an AI prompt, created in parallel in about 10 minutes for $2 total. The Thoughtful Minimalist. The Personal Story Teller. The Modern Luxury. Ten variations, zero developers hired.
I ended the post with a question: which one is best? But the real question I was asking myself was much bigger: what does this mean for what I do for a living?
The Background Nobody Tells You About
I should back up. I'm Poul Kjeldager Sørensen, and I've been building software professionally for over two decades. I work at Context& (formerly Delegate) in Denmark, where we help companies become more effective through IT. I've done everything from building software for ESA (the European Space Agency) to consulting on Power Platform projects. I'm not a hobbyist who discovered coding through ChatGPT — I'm a professional developer who discovered that the entire craft was about to transform.
That context matters because the story I'm about to tell you isn't "non-technical person discovers AI can code." It's "experienced developer discovers AI can code at a level that changes the economics of the entire profession."
The Christmas Rabbit Hole
What I found over that Christmas holiday was Bolt.new. And Lovable. And v0. And the Anthropic Console. Each one let you describe what you wanted in plain language and watch working software appear. For someone who'd spent 20+ years translating ideas into code through careful architecture and manual implementation, this was... disorienting.
I started with simple things. A landing page here. A prototype there. But the feedback loop was addictive — describe, generate, iterate, ship. By the time New Year's rolled around, I'd lost count of how many sessions I'd run. The tools weren't perfect. The code wasn't always clean. But the speed was something I'd never experienced.
What made it different from, say, using a template or a page builder was the flexibility. I wasn't constrained by someone else's layout. I wasn't dragging and dropping components into predetermined slots. I was describing what I wanted — with all the nuance of a professional developer's understanding — and getting back something that mostly worked.
The Moment It Clicked
There's a specific feeling when you've been building software for decades and you suddenly realize the tools have shifted under your feet. It's not excitement exactly — it's more like vertigo. The question "Could I have solved this 5 years ago with HTML templates and some string replace?" crossed my mind early. The answer was yes, technically. But "technically possible" and "actually practical" are separated by a canyon of effort.
What AI did was collapse that canyon. The cognitive load of going from idea to working prototype dropped from days to minutes. Not because the AI was smarter than me — it wasn't — but because it could do the tedious translation work at machine speed while I focused on what to build rather than how to build it.
By late December, I was generating A/B test variations at a cost of $2. The concept of paying developers to build 10 versions of something had always been absurd. Now it was trivial.
What I Didn't Know Yet
On December 23rd, 2024, I had no idea that two months later Andrej Karpathy would tweet about "vibe coding" and give a name to what I was doing. I didn't know that within six months I'd log over 1,000 coding prompts. I didn't know I'd end up building tools for building tools, running biweekly live coding sessions, or seriously contemplating whether I was coding myself out of a job.
I also didn't know about Cline — my first Cline prompt wouldn't come until February 4th. Or Claude Code, which would arrive that summer and change everything again. Or that I'd eventually discover multi-agent workflows and start thinking about software development as something that could be entirely autonomous.
All I knew on December 23rd was that I'd spent $2 and 10 minutes and had 10 websites that, individually, would have each taken a developer at least a day.
The rabbit hole was open. And I was already falling.
