How this site was made
I built this portfolio with Claude in October 2026. I directed it the way I direct any AI work: set the standard, review what comes back, and keep only what holds up.
What I decided
The direction, the case studies and every claim on the page. My rule for the writing was simple: no number I can't source, and no estimate I can't explain out loud.
I also decided what stays private. One case study covers a confidential client, so it carries no names and sits behind a password.
What Claude did
Claude drafted the page structure, the design and the code, and turned my notes into first drafts of the case studies. It also worked as an editor. For each draft it listed the details only I could confirm, and it flagged lines that said more about a client than they should.
The drafts went through several rounds. Each time, I answered the open questions, cut what I couldn't stand behind, and rewrote the parts that didn't sound like me.
How I work with AI
I treat AI as a collaborator. I bring the standard and the judgment; it brings speed, range and a second set of eyes, and it's expected to push back. On this site, Claude proposed structures and drafts, and I turned down the ones that didn't fit. It challenged me in return: asking where a number came from, or whether a detail said too much about a client.
The output is never the end of the work. Nothing here went up because a model wrote it. It went up because I read it, checked it against what I know, and decided it was true.
Why this page exists
The work I want to do next is directing AI well: writing the instructions, setting the evidence standard, and judging the output. This site is a small example of that work, so it seemed right to show how it was made.