Agentic AI Engineer - Senior Manager

EY•Hartford, DC

18d•$144,000 - $374,000•Hybrid

About The Position

EY is hiring a Senior Manager-level Agentic AI Engineer for Challenger, a program developing new AI-enabled services and the shared technology, methods, and risk and quality guardrails needed to build and scale them. The role focuses on engineering AI agents that carry work from user intent to verified outcomes, with reliable execution loops, tools, context, permissions, evaluations, and runtime controls. It combines applied AI, distributed systems, and product engineering, including long-running behavior, enterprise integrations, and production failure handling. The engineer will choose the appropriate balance of deterministic workflows and autonomy, while accounting for operational reliability and user experience. The position requires up to 50% travel and flexibility to work in an office in a hybrid capacity. The role includes designing agent execution models; building reliable, narrowly authorized tools; shaping planning, routing, retry, stopping, and handoff behavior; creating durable context; bounding agent authority; evaluating complete trajectories; and operating recoverable long-running execution. The engineer will make behavior observable, preserve user ability to intervene, and collaborate across product, platform, knowledge, security, and evaluation teams to develop reusable patterns and operational practices. The position requires a bachelor’s degree in a relevant field and 9–12+ years of relevant experience. EY lists base salary ranges by location: $172,800–$359,000 for New York City, Boston, and Washington DC Metro Areas, Washington State, and Southern California; $180,000–$374,000 for Bay Area California offices; and $144,000–$329,100 for other U.S. offices, including Sacramento. Compensation is determined by factors including education, experience, knowledge, skills, and geography. Benefits include medical and dental coverage, pension and 401(k) plans, and paid time off options, including flexible vacation and designated holidays, winter/summer breaks, and personal/family care leave.

Requirements

Bachelor’s degree in a relevant field of study.
9–12+ years of relevant experience.
Strong record in software or applied machine-learning engineering, including ownership of production systems and their live behavior.
Hands-on evidence of shipping large-language-model applications, agents, or complex model-driven workflows used for real work.
Strong programming capability in Python, TypeScript, or another language used for production services and integrations.
Experience with model APIs, structured outputs, retrieval, tool use, and evaluation of variable behavior.
Practical command of asynchronous work, state, concurrency, idempotency, retries, and failure recovery.
Experience implementing access boundaries and scoped authority for applications or automated workloads.
Evidence of improving live systems through traces, metrics, logs, incident analysis, and durable remediation.
Ability to reason about agent planning, tools, state, handoffs, and coordination, and to choose simpler workflows when they perform better.
Strong software engineering and distributed-systems judgment, including experience building dependable services under failure.
Ability to balance quality, latency, cost, and user control and explain technical trade-offs across disciplines.
Ability to travel up to 50% and flexibility to work in an office in a hybrid capacity.

Nice To Haves

Experience with durable agent execution, multi-agent systems, or human-agent collaboration.
Experience integrating agents with permission-sensitive business systems, browsers, or computer-use environments.
Experience with MCP or comparable standards, enterprise retrieval, knowledge graphs, or context services.
Experience with sandboxed execution, durable workflow engines, or systems requiring strong evidence and control.
Practical contributions to agent, evaluation, developer tooling, applied AI, or open-source communities.

Responsibilities

Choose the appropriate execution model, using deterministic steps, a single-agent loop, or coordinated agents according to measured performance.
Build dependable connectors and interfaces with clear contracts, validated outcomes, narrow authorization, and recoverable or escalatable errors.
Engineer agent planning, routing, delegation, stopping, retry, and handoff behavior across complete workflows.
Create durable context from instructions, user state, retrieval, memory, and environment feedback while preserving relevance, provenance, freshness, and access controls.
Bound agent authority through scoped identities, credentials, budgets, approval points, and emergency containment.
Evaluate complete trajectories with representative tests measuring verified task success, tool use, recovery, human takeover, and prohibited actions.
Operate long-running execution using durable state, checkpointing, safe retries, versioned releases, rollback, and graceful degradation.
Instrument end-to-end traces and replay to diagnose failures, identify patterns, and turn production evidence into improvements.
Design intervention points so people can inspect, interrupt, redirect, and safely resume agent work.
Partner with product, platform, knowledge, security, and evaluation teams and publish reusable agent patterns and operational lessons.