AI is in almost everything I make. Here's exactly how, and where it isn't.
Most people who list "AI" on a resume can't tell you where their judgment ends and the model's begins. I can, because I designed it that way on purpose.
Two ways I build
Build it myself.
When speed matters more than scale, I prototype directly: custom HTML apps, Copilot agents, project hubs, and this site, built with Claude Code. If I can show a stakeholder a working thing on Tuesday, we skip three weeks of debating a slide.
Partner to scale.
When something needs to run for thousands of people, I work with product and engineering. I bring the use case, the requirements, and the user testing. They bring the architecture. That's how 14 agents made it into production.
How I choose If it's a test, I build it. If it needs to last, we build it.
My build loop
Define the requirement.
I turn a business need into a precise brief in plain language.
Generate and review.
I check the output against specs, brand rules, and copy rules. Nothing ships unread.
Test in the real environment.
I verify it live, in SharePoint, Copilot, or the browser, then feed the results back.
Document what I learned.
Confirmed behaviors go into guides and maintenance docs so nobody has to rediscover them.
Then I repeat until it works. The AI recognition program's scoring app took 55+ versions, and that's not a failure rate. That's the method.
Who does what
- what to build and why
- strategy and prioritization
- stakeholder and legal decisions
- final copy and sign-off
- anything with a person's name on it
- first drafts
- code and scripts
- data cleanup
- consistency checks across documents
- pressure-testing my thinking
Guardrails I keep
I verify every number.
AI drafts. I check the facts before anything goes out.
I use approved tools for sensitive data.
Employee and client information only goes through enterprise-approved platforms.
I build accuracy rules into agents.
For example, my briefing agent is told never to infer ownership from attendance and never to invent deadlines.
I say how a thing was made.
When AI helped, I'll tell you how.
Tools Claude · Claude Code · Microsoft Copilot (including custom agents) · ChatGPT · enterprise AI platforms
AI didn't run these programs. I did, using AI.