Senior Product Manager

About the role

We're building an AI system that learns from how domain experts make decisions — and gets measurably better over time. The challenge isn't building another chatbot or copilot. It's designing a system that can extract structured intelligence from messy, real-world professional workflows, identify reliable patterns across many decisions, and surface that knowledge to the point where it's actually useful — without requiring anyone to change how they work.

You’ll be the founding partner on the pod embedded in a team of AI and software engineers. This isn't a conventional PM role. You're the person who owns the seam between humans and the system, runs a parallel build track so the pod isn't single-threaded on engineering, and holds us accountable to whether the engine is actually learning.

You’ll report to the Director of Product and interface regularly with the leadership team as the voice of the pod — communicating progress, surfacing blockers, and translating system performance into strategic implications.

• Sit alongside brokers. Watch them work. Try interaction patterns. Discover what "human in the loop" actually means in practice — not as a phrase, as a working interaction pattern.
• Ensure the team is solving the right problem. You are the interface between what brokers need and what the system can produce.
• Map workflows into tangible assets, advice units, and recommendations for brokers to act on.
• Design the feedback loops that allow domain experts to validate, correct, and enrich what the system surfaces — without adding friction to how they already work.
• Stand up solutions brokers can touch by mid-week. Without blocking on the Lead AI Engineer. You'll use Claude Code, Cursor, Lovable, v0, or whatever gets there fastest.
• Run the learning backlog. Track what we've learned about the engine, what we still don't know, and whether each iteration is closing the gap. This is interpretive work — reading instrumentation, forming hypotheses, feeding them into the next cycle.
• Build instrumentation with the engineering team to track whether the system's intelligence is actually improving. If it isn't, the answer is stop — and you should find that useful, not threatening.
• Author the metric system — north-star, driver, and guardrail metrics tied to unit economics (revenue per FTE, cost per case) — and hold the team accountable to it.
• Recruit and run the pilot. Pick the first users yourself, sell them on it, walk each one through the system 1:1, and convert their feedback into the next build cycle. Adoption is yours to earn, not GTM's to deliver.

• You’ve shipped AI products in production. Not pilots. Not evaluations. Real systems with real users — including at least one with a feedback loop where human corrections improved the system over time.
• You build. You have side projects. You have artifacts you've shipped yourself. You can walk us through code you've personally written. Your daily workflow includes Claude Code, Cursor, Lovable, v0, Replit, or similar.
• Systems thinking: you design workflows, feedback loops, not feature sets. You've thought seriously about how systems learn and improve over time, not just how they process inputs. You connect the dots across systems and workflows
• You’ve done 0-to-1. You've defined success criteria for things that have never been measured. You've shipped before requirements were clear. You found it energizing.
• Field research is a reflex. Dropped into a domain you don't know, your first move is to sit with the people who do the work — think ten practitioners shadowed and interviewed in your first two weeks, and an artifact they rate worth using by week three.
• Experience in regulated or complex professional domains — insurance, fintech, legal, healthcare, or similar fields where the signal is human judgment and the stakes of being wrong are real.
• Comfort with ambiguity: you can define and make progress on a problem when the requirements are incomplete and success metrics don't yet exist.
• You operate in writing. TL;
  
  DR-first updates, standup recaps that read as decision logs, and status reporting that separates shipped from in-flight from scheduled —…