Lead Product Manager, Agent Platform

Required

• You’re ruthless about priorities. Our full stack approach means there’s always more to do than people to do it, so you keep the team on the highest-ROI work and you’re comfortable saying no to good ideas that come at the wrong time.
• You’re comfortable in a heavily regulated market. Compliance is non-negotiable for us. We work with regulators to push the industry forward and get better outcomes for borrowers. We play by the rules, and we play to win.
• You care a lot about our customers and their borrowers, and you hold a high bar for them. We’ll do whatever it takes to make them successful on Pylon.
• You default to software over manual work. One of our guiding principles is that anything in the backend mortgage process done by a human is a bug. We can’t automate everything right away, so when we need a temporary workaround, you flag it, write it down, and push for the real fix.
• You’re good with people. You’ll work with internal teams who want different things, engineers on the customer side, and a long list of model and tooling vendors, and you know how to build real relationships and get results through them.
• You’re willing to get into the weeds. You can learn mortgage well enough to sound like a 20-year veteran, and you keep up with AI research like someone who actually builds with it.
• 4+ years as a product manager. An engineering background is a plus
• You’ve owned and scaled a developer platform, infrastructure product, or API used by several teams or outside customers
• You’ve shipped LLM-based agents or agent tooling to production and know your way around tool use, MCP, multi-agent orchestration, and structured outputs
• You know evals well, including golden datasets, evals, LLM-as-judge, and regression testing, and you can connect model quality to business results
• You understand retrieval, including RAG, embeddings, vector databases, chunking, hybrid search, and reranking
• You understand the tradeoffs between prompt and context engineering, model choice, and fine-tuning, and how each one affects cost, latency, and accuracy