Staff Product Manager, AI

Brex Brex · Fintech · New York, NY +2 · Product

Product Manager for the Brex Assistant, an AI agent designed to handle employee expense, spend, and travel tasks through a conversational interface. The role involves defining product strategy, agent capabilities, and ensuring the agent's reliability and trustworthiness in a financial context. This is an early-stage product with significant investment, requiring innovation in evaluation, prompting, and feedback loops.

What you'd actually do

  1. Own product direction for the Brex Assistant — the strategy, the roadmap, and the day-to-day calls on what ships.
  2. Define agent capabilities that take real actions for users, and translate messy real-world expense and spend workflows into reliable agent behaviors.
  3. Invent the "how": design the evals, prompting approaches, and feedback loops that make the agent dependable — and re-invent them as the underlying models and tooling shift.
  4. Get hands-on. Prototype, measure, read transcripts, and iterate quickly toward what actually works rather than what looks good in a demo.
  5. Talk to users constantly, and let real usage — not assumptions — drive the next iteration.

Skills

Required

  • Product management experience
  • Experience defining product strategy and roadmaps
  • Ability to translate complex workflows into agent behaviors
  • Experience designing evaluation frameworks and feedback loops for AI agents
  • User research and iteration skills
  • Understanding of AI/ML concepts, particularly in agentic systems
  • Ability to work hands-on and iterate quickly

Nice to have

  • Experience in fintech
  • Experience with conversational AI
  • Experience with prompt engineering
  • Experience with LLM observability

What the JD emphasized

  • own the Brex Assistant end to end
  • set the product direction and strategy
  • decide which capabilities to build and how they should behave
  • talk to users constantly
  • figure out what will actually help them
  • go as deep as it takes to make it real
  • shaping the agent's behavior
  • writing and iterating prompts
  • designing the evals that prove it works
  • digging into traces when it breaks
  • set the quality bar for an agent that real people trust with real money
  • invent much of that
  • ownership and pace of a startup
  • backing of an established company
  • define agent capabilities that take real actions for users
  • translate messy real-world expense and spend workflows into reliable agent behaviors
  • design the evals, prompting approaches, and feedback loops that make the agent dependable
  • re-invent them as the underlying models and tooling shift
  • iterate quickly toward what actually works
  • talk to users constantly
  • let real usage — not assumptions — drive the next iteration
  • set and defend a high quality bar for an agent operating in a trust-critical, financial context

Other signals

  • AI agent for expense and spend management
  • Focus on conversational interface and action-taking capabilities
  • Early-stage product with significant investment and backing