Building AI with AI

During the agency years I had a team of six developers overseas. It was hard to manage. They were slow, and everything was manual.

So I built a boilerplate. Then I needed a CMS, and it started adding up.

In my manifesto I put it this way: “We built ‘agencies’ that were just jobs with more overhead. We scaled our revenue, but we scaled our stress.”

WebCore’s first commit, on 25 June 2024, was a referral page for MTS Moving.

I spent 18 months and $180,000 with a dev team building its foundation: the infrastructure, the load-bearing walls.

AI came in the moment I saw Vercel’s AI SDK. At first I built everything by hand. I set up RAG systems, and my plan was to create agents on demand for a whole web development team.

Building AI by hand was a terrible idea, and another expensive mistake.

I lost a great deal of cash flow trying to keep that team, until I couldn’t afford it anymore.

So I pivoted to building AI with AI, and that was a true differentiator.

I finished WebCore myself, working with Claude, without hiring more developers and without writing more code by hand. In the three weeks before launch, Claude co-wrote 659 changes.

I’m done building by hand. I’m done paying for effort. From now on, I only pay for outcomes.

Since 2016 I’ve run the agency-style work, and now I build the AI meant to replace it. I’ve lived every side of it.

WebCore went live on 1 October 2026. It is an AI for revenue for small businesses. It builds a fast website and keeps it fast, answers visitors when the owner is busy, and saves every enquiry as a lead. Owners ask for changes in plain words, the way they would ask a person.

I told my team MCP would be the standard. It is, and any AI agent can now connect to a WebCore site.

WebCore also publishes the Ontario Trades Web Index, monthly research done by Bro, the AI agent I built to study this market. And teachers get a free Lessons skill, which grew out of ClassMaven, my earlier platform for teachers.