Senior AI Product Manager; AI Builder

About The Role

WEX is putting governed, production-grade AI agents to work across Mobility, Payments, and Benefits to turn enterprise data, systems, and deep domain expertise into resolved outcomes at scale.

• Build to Learn:
  Put working concepts directly into users' hands early rather than waiting for perfect requirements.
• Prototype, Not Just Describe:
  Use AI tools, low-code/no-code platforms, and lightweight code/configurations (Markdown, YAML, JSON) to make product ideas tangible.
• Get Inside the Agent:
  Inspect prompts, system context, tool calls, traces, outputs, and failure modes instead of treating AI as a black box.
• Write and Run Evals:
  Define golden datasets, representative edge cases, scoring rubrics, and judge criteria to drive product decisions through evidence.
• Reimagine Workflows:
  Focus on end-to-end outcomes—determining where AI should reason, retrieve, automate, or route control to a human (and knowing when deterministic software or human judgment is safer).
• Drive Measurable Value:
  Measure success by concrete business outcomes (cost, cycle time, accuracy, revenue, risk, adoption) rather than output volume.
• Build for Scale and Reuse:
  Package learnings, prompts, evals, and agent patterns so every build makes the next deployment faster and cheaper.

• Embed directly with operators, customers, and domain experts across Mobility, Payments, and Benefits to map actual current-state workflows, friction, and failure demand.
• Define future-state experiences, set baseline performance metrics (cycle time, manual hours, containment, error rates), and select the smallest high-value slice to test feasibility and trust.

• Rapidly build and test agent behavior, system context, prompts, tool flows, retrieval/grounding, and multi-agent or human-in-the-loop orchestrations using WEX’s Agentic AI platform.
• Pair directly with your Tech Builder to open repositories, adjust configurations, test scenarios, examine execution traces, and debug agent logic together.
• Distinguish between rapid exploratory learning prototypes and production systems requiring engineering hardening.

• Treat evaluation as a core product discipline by defining representative, edge-case, and adversarial scenarios alongside domain experts.
• Establish pass/fail criteria, assertions, and scoring rubrics covering accuracy, tool selection, citation quality, policy adherence, hallucination rates, PII/PHI/PCI safety, and cost per outcome.
• Analyze failed traces to build error taxonomies (distinguishing between prompt, context, retrieval, model, or UX issues) to prevent "demo-driven development."

• Translate learnings into versioned repository artifacts (PRDs, acceptance criteria, decision logic, eval harnesses) that serve as a shared source of truth.
• Maintain clear traceability:
  User Problem → Workflow → Requirement → Scenario → Eval → Production Metric.
• Navigate enterprise legacy systems, fragmented APIs, and complex operational policies to show working software in real environments.

• Partner with Risk, Compliance, Security, Legal, and AI Governance from day one to embed decision boundaries, entitlements, auditability, prompt-injection defenses, and fallback controls.
• Instrument live product usage to connect agent performance directly to top-line and bottom-line business ROI.
• Package reusable patterns, runbooks, and controls so domain teams can seamlessly operate and extend capabilities as solutions mature.

• 7+ years of Product Management experience shipping software, data, platform, automation, or AI…