Builder PM who ships AI features, writes the code, and builds the agents that ship the next one

Ask FDE (RAG-powered Q&A over specs and meetings, with source citations), Jira ticket generation, and weekly update drafting. I owned the prompt design and the user-side behavior (when AI fires, how outputs surface, accept/edit/dismiss); engineering owned the data-ingestion rails. I was also PM on two more AI features in the platform - meeting matching and spec generation - where I owned requirements and testing while engineering owned the build. Each feature came from a bottleneck I hit during 34 manual onboardings. Each one keeps a human in the loop.

Combined usage across the three features I shipped grew ~51x in 6 months - from 7 events in January 2026 to 357 by 06/26/26, with 4 days still pending in the month. Ask FDE ramped from 2 events in January to 308 in April across 31 unique analysts. The weekly-update feature went from 24 events in its launch fortnight to 245 a month later.

A standalone AI agent that cuts app-store screenshot generation from 2–3 hours to 15 minutes. One command: log in, screenshot 6 pages at 3 device sizes, composite into store-ready frames, create a Jira ticket. Flagged internally as a model AI-adoption win by the team driving AI rollout at Cut+Dry.

A team-facing PM agent that turns Slack chatter and Jira tickets into specced work for the FDE Platform. It subscribes to the team's Slack channels and a Jira webhook, self-classifies whether each message is a real work request (silent if not, even when @mentioned), grounds in current product behavior, runs one clarifying question at a time in-thread, and proposes 2–3 approaches with PM-language trade-offs. Hard gate: it only writes the Specification Document after explicit human approval. Then it links the spec on Jira, transitions the ticket to PRD, and stays in-thread for follow-ups. Never writes code, opens PRs, or merges.