How I run a 70-project AI portfolio
Every AI project gets a portfolio owner, an executive stakeholder and a path from first version to rollout. My job is keeping all of it in view.
AI projects I’ve created, at every scale
roles I play: collaborator, support or primary builder
months from one department to every department
Portfolios follow how the company is organized
Creating projects is one of the most important parts of what I do, because the whole company is shifting how it works around AI. A portfolio is either an entire department or a subset of one. Finance might have an underwriting analyst portfolio rather than one bucket for everything financial. Marketing has a portfolio for each of its practice groups.
Every project has two people on the hook
Each project belongs to a portfolio, and each portfolio has a portfolio owner and an executive stakeholder. Those two people decide when something gets deployed and what the action items are, so nothing ships without the people who own the outcome.
From spec to rollout
- Define
I define each project in Linear as deeply as I can.
- Milestones
Break it into milestones toward a minimum viable product.
- Test group
The first version goes to a testing group.
- Review
It’s tested with the portfolio owner and the executive stakeholder.
- Deploy
Then it’s released to the audience it was built for.
Contextual awareness across everything
Everybody is building now, which is the point. My job is to keep contextual awareness of all of it: no duplicate effort, no overlap, and when two teams are building the same thing, working out how to merge it.
On any given project I’m in one of three roles: a collaborator, exclusive support, or the primary builder. Across the portfolio I’m usually doing all three at once, while keeping watch over everything else that’s being deployed.
The tools
We track everything in Linear, because it integrates directly with our engineering team. I connect Claude to Linear and to the other third-party software we use through MCP, so the tracking and the work stay in the same place. I’ve also used Wrike and Jira, and plain Excel, which is simple and effective.
How it started
My title has been AI Process Engineer from day one. I started in one department, and once that department was using AI, I went to every department. That took four months.