AI that holds up in production
I treat agents like any critical system: engineered context, layered guardrails, challenge agents that check the work, and evaluation against real human decisions on production traffic instead of a test set the prompt was tuned to. In trial, the AI agreed with human reviewers 98.73% of the time.
Product judgment, not just tickets
When there's no PM, I go to the people doing the work. I've shadowed reviewers in Seattle, Austin, and Costa Rica, written requirements with program teams, and built working prototypes so leaders could decide with something real in front of them.
Scale, reliability, and cost
I've run checkout systems at 10,000+ TPS across 10+ countries and taken a critical API from 0.45 to 83.6 TPS ahead of re:Invent. I migrate without drama: parallel runs, feature flags, zero downtime, and bills that go down.
Leadership that multiplies
I've led up to three teams at once (30 engineers) across 5+ orgs, and I work directly with directors and present to VPs. I mentor about three people at a time. Mentees include 5+ engineers promoted, two promoted to Senior, and a content reviewer who moved into a technical role.