Lead Product Manager, Agent Platform
Lead PM owning Pylon's Agent Platform: building internal tools to build, test, ship, and monitor LLM agents, plus external APIs/MCP servers so customer agents can do mortgage work. Requires 4+ years in product management, hands-on coding, and deep LLM/agent experience; based in NYC or Palo Alto, in office.
- Compensation
- Not disclosed
- Location
- NYC or Palo Alto (in office)
- Work setup
- Onsite
- Seniority
- Lead
- Employment
- Full time
- Experience
- 4+ years
- Category
- Product
- Posted
Skills
Must have
- 4+ years as a product manager; engineering background is a plus
- Owned and scaled a developer platform, infrastructure product, or API used by several teams or outside customers
- Shipped LLM-based agents or agent tooling to production; familiar with tool use, MCP, multi-agent orchestration, and structured outputs
- Deep knowledge of evals: golden datasets, LLM-as-judge, regression testing; can connect model quality to business results
- Understands retrieval: RAG, embeddings, vector databases, chunking, hybrid search, reranking
- Understands tradeoffs between prompt/context engineering, model choice, and fine-tuning (cost, latency, accuracy)
- Familiar with tracing, guardrails, prompt injection, and protecting sensitive data like borrower PII
- Writes code and can commit to a production codebase, build prototypes, and go deep on architecture with engineers
- Has built 0-1 enterprise software in a regulated industry
- Can write SQL and define metrics for agent and platform usage
- Must live in or be willing to relocate to Palo Alto (CA) or New York City
Nice to have
- Mortgage or fintech infrastructure experience, or work on document understanding (OCR, extraction, classification)
Full description
About Pylon
America’s $13T mortgage market is one of the most important financial systems in the world. It underwrites the middle class and is the mechanism through which millions of families build wealth. But while every other financial instrument has been simplified to an API call, mortgages are still assembled by hand.
We started from zero and created the first vertically integrated mortgage platform that turns origination into a single API.
Publicly traded companies and the country’s largest originators are already building on Pylon. Revenue is compounding monthly. We’re backed by Peter Thiel, Conversion Capital, QED, Citi, Fifth Wall, and the founders of Ramp, Blend, and Mercury.
Working at Pylon isn’t for those seeking comfort. The people who thrive here have high agency, strong opinions, and a track record of delivering outcomes without direction. Many of us are former founders. We move quickly, challenge each other directly, and take full ownership of results. It’s hard work, but it will be worth it.
Join us in building America’s mortgage rails.
About the role
Agents are starting to do a lot of the work in mortgage, both at Pylon and at our customers. We’re hiring a Lead PM to own the Agent Platform, which is what every Pylon agent is built on and how outside agents work with Pylon.
The job has two halves. The first is internal. You’ll build the tools and infrastructure our engineers, operators, and business teams use to build, test, ship, and monitor agents, for everything from reading documents to underwriting to closing.
The second is external. Our customers are building their own agents, and you’ll make Pylon the easiest place for those agents to do real work. That means MCP servers, APIs designed with agents in mind, and the permissions and guardrails that keep it all safe.
What you decide here will shape how quickly we can automate the mortgage process, and how easily the rest of the industry can build on us.
What you’ll do
- Set the direction for agents across Pylon and get product, engineering, and operations aligned on where they’ll make the biggest difference
- Own the roadmap for the platform’s core pieces, like tools, orchestration, context and memory, retrieval, evals, observability, and guardrails
- Make it much faster for internal teams to go from an idea to an agent running in production, and give them shared patterns so nobody builds the same thing twice
- Treat evals as a product of their own. Teams should have shared datasets, benchmarks, and regression checks, plus human review wherever a regulated decision needs it
- Own how external agents work with Pylon, including MCP servers, APIs, scoped permissions, rate limits, and docs that both people and LLMs can follow