Senior Technical Product Manager – Agent Evaluation & Observability

American Express•New York, NY

About The Position

The Senior Technical Product Manager will lead the roadmap for shared tools and workflows that help teams test and improve agent-enabled customer experiences. The role owns product roadmap, requirements, launch, adoption, and ongoing improvement of evaluation and observability capabilities. It works with engineering, enterprise platform teams, and domain product owners to help teams assess agent performance and determine what to improve; domain product owners remain accountable for release decisions. The position defines workflows for evaluating task completion, diagnosing failures, and improving agent experiences. It connects evaluation with operational evidence, establishes meaningful measures, and develops reusable product capabilities while incorporating privacy, data governance, uncertainty, and release-risk considerations.

Requirements

6–8+ years of experience in product management, product development, software engineering, technical platforms, analytics, quality tooling, or AI products.
Experience owning evaluation, testing, observability, analytics, or related technical products through launch, adoption, and iteration, with measurable user or business outcomes.
Practical understanding of AI or agent-enabled systems, including nondeterministic behavior, tool or API interactions, and the need for both automated tests and human judgment.
Ability to define metrics, interpret evidence, and identify sampling, coverage, or measurement limitations.
Familiarity with traces, logs, APIs, and technical troubleshooting.
Strong product discovery, roadmap, and prioritization skills, including translating ambiguous quality problems into usable tools and repeatable workflows.
Cross-functional leadership and clear communication, with the ability to balance customer outcomes, engineering constraints, and privacy or risk requirements.
Employment eligibility to work with American Express in the United States; the company will not pursue visa sponsorship for these positions.

Nice To Haves

Experience with LLM or agent evaluation, scenario design, human review, model-assisted scoring, or production failure analysis.
Familiarity with MCP, function calling, telemetry standards, or evaluation frameworks; ability to use code or queries to investigate evidence.
Experience building shared capabilities adopted by multiple teams or delivering products in financial services or another regulated industry.
Bachelor’s degree or equivalent experience.

Responsibilities

Own the roadmap for shared agent evaluation and observability capabilities, from user discovery and requirements through launch, adoption, and improvement.
Partner with domain owners to define representative journeys, prompt sets, expected tool behavior and outputs, scoring rubrics, and acceptance criteria.
Define requirements for automated evaluation, human review, and tools that help teams diagnose failures.
Define product requirements for traces, events, diagnostic views, and review tools that connect customer intent, tool or API calls, and observed outcomes where data access permits.
Work with engineering, analytics, and existing enterprise evaluation and observability teams to extend shared capabilities, reuse instrumentation, and standardize evidence.
Identify where limited access to partner data prevents a complete assessment.
Partner with Identity and domain owners to test policy-relevant behavior, including denied actions, revoked access, and required approvals.
Define meaningful measures such as task completion, correct tool use, response quality, failure and recovery rates, latency, and cost.
Define how testing and monitoring work together across controlled test environments, end-to-end partner integrations, and production; clarify what each can measure and use findings to prioritize improvements with responsible teams.
Measure adoption and usefulness of evaluation workflows, including review effort, coverage, and time to identify problems.
Communicate the limits of findings based on sampled data.
Translate measurement gaps into a prioritized backlog, clear requirements, and actionable acceptance criteria.
Recommend when to extend existing tools, integrate a solution, or build a missing capability.
Build privacy and data governance into evaluation and telemetry requirements, including permitted uses of partner-provided data, appropriate access, minimization, and retention.
Communicate uncertainty, release risks, and tradeoffs clearly.