Because it is healthcare, compliance comes first — not as a footnote, as the first constraint. Every third party that touches protected health information sits under a signed BAA, and anything without one is kept PHI-free by design. When PHI and Claude meet, it is Claude through AWS Bedrock under a BAA, never the consumer app. I will make a business decision that costs us before I cut a compliance corner. Honestly, I am a little proud of how many bad-for-business decisions we make in the name of doing the right thing — compliance, morality, and doing right by people, in that order, every time.

Claude does a real share of the work. I use Claude Code daily, connected through MCP to the tools the business actually runs on: ClickUp, Notion, Slack, Gmail, Calendar, Make.com, Fireflies, Canva, Gamma, and Semrush. Not as a demo. As the way things get done.

Claude is my main driver, but it is not the only model I pay for. I keep Pro memberships to OpenAI's ChatGPT, Google's Gemini, and Perplexity, and I run local 30B models on Ubuntu through Ollama. The reason is plain: I want to know first-hand where each model is actually stronger. Most days I open four of them across two monitors and put the same question to all four at once — a kind of standing group of friends, four perspectives on one prompt. Then I read the answers side by side and study where they diverge. It is fascinating, and over time it has given me a real edge on which model to reach for which use case. Claude wins most of those comparisons, which is exactly why it is the main driver — I know that because I checked, not because I assumed.

That habit shows up in the year-end numbers. My 2025 ChatGPT recap landed me in the top one percent of users by messages sent, and GoHighLevel ranked me in the top one percent of its AI users globally. Both came from personal, hands-on use — my own prompts, my own studying of the responses. None of that volume was farmed out to external agents running on my behalf; it is one person, four models, two monitors, asking better questions.

I document this publicly because the gap between Claude content and Claude implementation is enormous. Plenty of people can list what the model could do. Very few show what they built with it, how they built it, and what it cost them in rework. I would rather be useful than impressive.

The honest version of how I work: no shortage of ideas, a lot of controlled chaos, and a fair amount of ADHD. The ideas are never the bottleneck — discipline is. I am lucky to have people around me who can hear 'I have a hundred and fifty ideas' and ask the only question that matters: which three to five are we actually shipping in the next ninety days? This entire site is about that part — execution, not intention. I do not announce resolutions in November to start in January. I just start. Talk is cheap; the build is the receipt.

Everything on this site ran in my own business first. If it is in a Build Log, it shipped. If it is in a Playbook, it is the process I still use. When something stops working, I update the page or mark it dead.