The site you’re reading was built by me and Claude, working together. Not “generated.” Built. There’s a difference, and it’s the whole point.
I’m telling you because hiding it would be daft, and a bit dishonest. Plenty of people in my line of work are using these tools and hoping no one asks. I’d rather show you what it looks like, because it isn’t what most people imagine.
What working with AI looks like
The setup is plain. I run Claude Code, which works inside my terminal, right alongside the live website. It reads the real code, runs commands, and deploys changes to the server, with me directing and checking every step. No toy chat window, no copy-paste.
It isn’t me typing “make me a website” and walking away. It’s closer to having a very fast, very well-read junior who never tires, with a senior who knows what “good” means steering every decision. I bring the judgement: what to build, what matters, what’s wrong, when something that looks right isn’t. Claude brings the legwork, at a speed I can’t match alone.
An example, from this site, this week.
The homepage wouldn’t hit a top performance score, and it has no images on it, which made no sense. The score kept bouncing between runs, too. The obvious suspect was an animated terminal panel in the header. We measured it instead of guessing, and the suspect was innocent. The cause was the main heading: the browser painted it once in a stand-in font, then repainted it a fraction of a second later when the proper font loaded, and that second paint wrecked the score.
The fix was niche: measuring that the real font sits at 83.8% of the width of the stand-in, then matching them so the swap becomes invisible. Score fixed. It stopped bouncing. The change itself took ten minutes; finding it was the job.
I’ll be straight about what did the work. Claude didn’t know the answer; it proposed wrong ones too. We got there by knowing how to measure, what to rule out, and how to tell a real solution from a plausible-sounding one. That’s the judgement. The legwork, pulling the data, testing the theory, writing and shipping the change, happened in a fraction of the time it would have taken me alone. That’s Claude. Neither half gets there as fast without the other.
The uncomfortable bit
I won’t dress this up: if you build for the web and you’re not learning to work this way, you’ll get left behind, or replaced. Not by AI on its own, by the people who’ve learned to use it well. The gap between those two groups is widening fast, and it’s already big.
That’s not me being gloomy. I’m energised by it. Problems that used to eat a day take an hour. I can take on work I’d have turned down two years ago. The craft hasn’t disappeared, it’s moved. It’s less about how fast you type and more about knowing what’s worth building, and telling good from nearly-good.
For you, as a client: senior judgement and the leverage of these tools, from one person, without an agency’s overheads. Faster, cheaper, and, done right, better. The “done right” is the human part. It always will be.
That’s why I’m telling you how the site was built. It isn’t a confession. It’s the pitch.
One more honest thing: this post was written by Claude too. I shaped it, edited it, and put my name to it, and every word reflects what I think, but the draft came from the same partnership I’ve described, Claude Code in my terminal. That’s the point. Like I said: make use of these tools, don’t hide behind them.
This way of working is now something you can buy: the AI-Search Readiness Audit, the Forensic Ads Audit and the AI Operations retainer are all delivered exactly as described here — machines doing the volume, me doing the judgement.