Who are we?
Equinix is the world’s digital infrastructure company®, shortening the path to connectivity to enable the innovations that enrich our work, life and planet.
A place where bold ideas are welcomed, human connection is valued, and everyone has the opportunity to shape their future.
A career at Equinix means being at the center of shaping what comes next and amplifying customer value through innovation and impact. You’ll work across teams, influence key decisions, and help shape the path forward. You’ll find belonging, purpose, and a team that welcomes you—because when you feel valued, you’re empowered to do your best work.
Job Summary
We’re looking for a Principal Product Manager to own Agent Intelligence: the evaluation, observability and feedback systems that tell us how well our agents perform, why they fail, and whether each change makes them better. You will set the quality bar for agents across Equinix and build the platform that enforces it.
This role is for someone who has done this before at scale. You have built or run evaluation or observability for production ML or LLM systems with many moving parts. You came up through systems engineering, and you can go from a dashboard to a trace to a root cause yourself. Now you want to apply that to agents working on real enterprise data, at global scale, where being wrong has real consequences.
Responsibilities
Set the quality bar
Define what “good” means for each agent and for the system end to end. That covers correctness, grounding in sources, completeness, tool-use accuracy, workflow adherence, latency, cost and business outcome
Turn product requirements and acceptance criteria into measurable evals, each with a clear threshold and owner
Set scoring standards and rubrics that hold across agents, workflows, regions and industries, and that both experts and automated judges apply consistently
Build the evaluation platform
Own the roadmap for offline and online evaluation. This includes golden datasets, scenario and regression suites, LLM-as-judge pipelines calibrated against human experts, and human-in-the-loop review
Make evaluation part of how agents ship, with eval-based quality gates in CI/CD. No model, prompt, tool or data change reaches users without passing agreed thresholds
Measure reliability, not just accuracy, using repeated trials, consistency across runs, confidence intervals and confidence calibration
Evaluate multi-component systems
Evaluate each component and the whole system: retrieval, entity resolution, ranking, tool and MCP calls, handoffs between agents, and final outputs
Build trajectory- and trace-level evaluation that traces an end-to-end failure back to the component that caused it, and test how errors spread across handoffs
Define the contracts between components (schemas, provenance, confidence) so quality can be checked at every boundary
Make agent behavior observable
Partner with engineering on tracing and telemetry across agents, tools and data sources, so every output can be traced to its inputs, sources and decisions
Build monitoring for drift in inputs, data sources, models and tool behavior, with alerts that catch regressions before users do
Give leaders and product teams a trusted view of quality, reliability, cost and adoption by agent, workflow and segment
Turn real use into improvement
Mine production traces, user edits, dismissals, ratings and expert audits for recurring failure modes, edge cases and data gaps
Turn every confirmed failure into a regression test, a product requirement, or training and tuning data, and track it through to a fix
Run experiments and A/B tests that show whether changes improve accuracy, reliability, adoption and business outcomes
Protect trust
Build evals and controls that enforce data boundaries. Internal data stays internal, customer-facing content uses only approved material, and confidential details never cross between accounts
Partner with security, legal and data governance to meet regulatory…