AI product manager building agents people can trust with real systems.

I'm an AI/ML product manager on U.S. Bank's enterprise data platform, with six years in technology. I joined as a software engineer in November 2022 and still build close to the system: agents that work across our tools, the evaluations that check them, and the fixes when production breaks. I also sit on the customer's side of the platforms we buy, from security reviews to escalations with the vendor's engineers.

Outside work, I build small products and agents of my own, and I publish what I find.

<aside> ✉️

Get in touch

Email · LinkedIn · Substack

Try them: PM Liftoff · Stack Spoon sample

</aside>


What I work on

Three questions run through everything on this page. Each tab lists the work that answers it.

Connect

<aside> 🔌

Can an agent reach the right systems with only the access it needs?

</aside>

Test

<aside> 🧪

Can people see what it will change before anything goes live?

</aside>

Observe

<aside> 🔭

When it runs, or breaks, can we see why and fix it fast?

</aside>


Selected work

Things I built

Writing

I also keep a prompt library of the prompts I reuse.


How I work

<aside> 🛡️

Accountability

"AI proposes, humans dispose, and the boundary between them has to be visible."

My agents stop at a human gate: the deployment agent at an engineer's sign-off and the manual apply, Virtual Engineer at merge-request approval.

</aside>

<aside> 🔓

Access

"AI for all. Non-technical people need to be empowered, not left behind."

The retrieval assistant, the office hours and PM Liftoff each take an expert out of someone's path.

</aside>

The pattern my agents follow

The pattern my agents follow: connect, test, observe

The pattern my agents follow: connect, test, observe

Both agents follow this pattern, and it maps to the three questions at the top of this page.


Background