I Hid My AI Coding for a Year. Then the World Caught Up.
Despite my medical background, I've been building software with AI since the first ChatGPT release. I told almost no one.
Discussion didn't seem like a reasonable or exciting option. Sentiment online was confident:
"It just writes boilerplate."
"You'll never build anything real or usable."
"It's a toy for demos, not production code."
Months before Andrej Karpathy coined 'vibe coding', SWEs online were already forecasting how limited AI would be at any point in the near future.
So I kept quiet, and kept building.
There was a hidden benefit to learning to code with AI from its public conception. I learnt with it as it grew up.
I'm a doctor, so unsurprisingly 18 months ago I couldn't tell you what React was. I didn't know how to use git, but I had done enough Codecademy to print 'Hello, world!'.
I had one advantage over every senior developer dismissing AI code: I didn't understand how much progress was needed. No idea about the "proper" way to do things. I just saw what worked, and I saw it getting better fast.
Anyone unimpressed missed what I could see. I saw exactly how far the 'edge' of AI coding could take you, and how far it moved in just weeks and months.
My ignorance was really my motivation safety net. I focused purely on outcomes and tests.
Every few months, something that was impossible became possible. Features I'd shelved as "too complex" suddenly weren't. The context window grew. The reasoning improved. My ambitions grew with it.
While some coders still debated whether AI could really code something for production, I was building.
What I Actually Built
Aestheticc started as a wrapper for text and image generation — helping aesthetic clinics with marketing content. Basic stuff.
Then it became a CRM. Clients, appointments, treatment records.
Then it got complicated. Inventory tracking tied to appointments. Automated billing. Reminder sequences. Consent forms. The kind of relational data model where everything touches everything else.
A year ago, I couldn't have built this. Neither the AI nor I had enough context. The models couldn't hold the full system in their head and neither could I.
Then, last month, I ported the entire web app to iOS.
It took two weeks.
A year ago? Six months minimum. Probably longer. Probably never, honestly — I would have quit.
The Window Is Closing
The gap is widening, and it's widening fast.
Some people using AI for code are still on web interfaces. ChatGPT in a browser tab. Maybe Claude. They give it a prompt, wait 20 seconds, copy-paste the output, hit errors, go back and forth.
That's not how this works anymore.
Even vibe coders live in an IDE or the command line.
I use Claude Code in my terminal. It reads my entire codebase. It runs for hours — three, sometimes six — autonomously making changes, running tests, fixing what breaks. Some people in the industry are running 20-hour tasks.
That's not a 2x difference. That's 100x. Maybe 1000x.
And every 3-6 months, the gap gets wider.
If your competitor starts using this tomorrow and you wait three months, you're not three months behind. You're in a different category entirely. You cannot catch up through effort. The leverage is simply too asymmetric.
What Changed
Suddenly, Linus Torvalds was vibe coding. Guillermo Rauch was talking about a hundred million builders. The conversation had moved from 'will it ever work' to 'how do we use it.'
But here's the uncomfortable truth: if you're just starting now because Linus has, you're already behind the people who started when it felt weird. When everyone said it would never work.
The Actual Advice
If you're a doctor, a lawyer, someone in finance, someone who's thought "I wish I could build something but I'm not technical" — here's what you need to know:
The window where AI is roughly human-level at coding is months, not years. We're in a strange moment where you can still learn alongside it, where your growth and its growth can sync up.
That moment is ending.
Not because AI will stop improving. Because it will improve so fast that "learning with it" stops being possible. It will just 'do the thing', and you'll either know how to direct it or you won't.
Start now.
Not because you'll definitely build a company. But because knowing how to work with AI is going to matter for every part of the economy, and the people who figured it out early will have a compounding advantage that late starters can't overcome.
I'm a doctor who couldn't use git, and 'built' an iOS app in a couple of weeks over Christmas two years later.
The window is open, but not for much longer.
I'm Shane. I left the NHS to build software for aesthetic clinics. I'm still figuring it out, one Claude session at a time.