The confession
I build software for a living. For 30 years I’ve shipped systems for other people — rescuing the troubled ones, modernizing the old ones, building the new ones from scratch. And for about three years, our own website sat there, dated and thin, while I did nothing about it.
It wasn’t a mystery why. I knew exactly what I wanted it to say. The trouble was finding the room to do it. The hours I had went where they should — to clients, and to keeping up with technology that changes under you constantly, one new release and training video after another. What little time was left, I couldn’t bring myself to spend on the slow, finicky middle: the keyword research, the competitor digging, the design system, the re-platforming, wiring up the contact form. None of it hard. All of it tedious. So it waited. And waited.
Then, over one weekend, I rebuilt the whole thing. This is the honest account of what that took — and, just as important, what it didn’t.
What I actually did that weekend
I worked with Claude, and ran a lot of the writing through ChatGPT for a second opinion — one model to draft, another to pressure-test. Across two days:
- A full competitive analysis. Scanned a dozen competitors, pulled their positioning, services, trust signals, and content, and scored a matrix of where we genuinely win and where we had gaps to close. Work that would normally eat a week.
- The SEO strategy, from scratch. Keyword research, clustered into themes, mapped to a real page architecture instead of a single brochure page.
- A complete Figma design system. Color and type tokens, a spacing scale, components — then carried consistently through every page of the build, so the site feels like one thing instead of ten.
- A full re-platform. Rebuilt on a modern, fast static stack and deployed to inexpensive, commercial-friendly hosting.
- A real Salesforce integration. And that last one is the part I want to dwell on.
The part that wasn’t a toy
When someone says they built a site “in a weekend,” you picture a pretty brochure. So here’s the unglamorous part that proves it wasn’t.
The contact form doesn’t just email me. It creates a lead in Salesforce, server-side, with the secrets handled properly and never exposed to the browser. It captures first-touch attribution — where the visitor came from, the campaign, the landing page, the referrer — and tags each lead by which service the person was looking at, so the routing is clean on the other end. It’s protected against spam without making a human prove they’re a human. And the whole thing is designed to feed a CRM we actually intend to use, not to sit in an inbox.
That’s systems work — the kind of integration that usually carries its own project plan. Doing it in the same weekend as the redesign is the real headline, because it’s the evidence that speed didn’t come at the cost of depth.
Where the judgment mattered
Here’s the part the breathless “AI changes everything” takes leave out: the tool was confidently wrong, often enough that without 30 years of scar tissue I’d have shipped the mistakes.
It kept over-indexing our positioning on a single technology stack until I corrected it — our actual strength is that we work across Java, .NET, and iOS, and the copy had to say so. It proposed architectural shortcuts that I knew would bite us in a year. Its first drafts of the copy were stuffed with the exact clichés we make fun of. Every one of those was a call that came from experience, not from the model.
The pattern was consistent all weekend: the AI removed the friction, and my experience supplied the quality bar — the taste to know when the output was wrong, when to override it, and when to throw work away and start again. The speed was real. So was the judgment that made the speed safe.
Why this isn’t “AI can do anyone’s job now”
If anything, the weekend convinced me of the opposite. The value wasn’t that an AI wrote code and copy. It’s that a senior person with judgment, plus the tool, is a genuine force multiplier. The years of taste are exactly what made the output good — strip those out and you get a fast pile of plausible-looking mistakes.
It’s the same point I made in Don’t shoot the lamp: the win isn’t using a powerful tool for everything, it’s using it for what it’s genuinely great at — and knowing the difference. And it’s the thing we tell clients every week, which I wrote about in Quality. Fast.: quality and speed aren’t a tradeoff. Done right, quality is speed. It was nice to prove it on ourselves for once.
What it means for the work we do
We didn’t just talk about modern delivery this weekend; we did it to our own house. And the way I worked — senior judgment steering AI to move faster without lowering the bar — is exactly how we think about AI on a client’s project. Not juniors hiding behind a model. Experienced people using it to get more done, and catching the things it gets wrong.
The website is the proof of concept. Your project would get the same discipline.