Moving AI agents from a playground demo to production requires serious software engineering discipline. We are looking for an AI Agent Engineer who knows how to build scalable, secure, and fully auditable agentic systems.

In this role, you won't just tweak prompts until they "feel right." You will build spec-driven orchestrator and sub-agent architectures, manage sensitive data classifications, integrate tool gateways, and establish comprehensive test cases. You’ll work closely with Product, Security, and Cloud Architecture to ensure every agent operates strictly within its designated guardrails and risk tiers.

Skills and Competencies:

• Write full agent specifications — purpose, capabilities, input/output contracts, guardrails, and tool boundaries — to the same level of rigor a safety-critical system would get, not a quick README.
• Design and build orchestrator agents that hand work off to specialized sub-agents, and hold the line on a simple rule: the orchestrator coordinates, it doesn't do the work itself.
• Write and test system prompts, and design tool schemas that stay accurate even when an agent has to choose between fifteen or twenty different tools — not just three.
• Work with the Data Steward to classify how sensitive each agent's data is, and with the Product Owner/Architect to set its risk tier, before anything goes up for registration.
• For each capability an agent has, decide whether it should be deterministic logic or model judgment, and build test cases that prove it actually meets that bar.
• Build every agent against the platform's tool gateway, so nothing an agent does falls outside what it's registered and approved to touch.
• Design for scale from day one — sessions that stay isolated from each other, failures that are predictable, and no assumptions that only hold up in a single test run.
• Prepare the evidence a reviewer needs before approving an agent for production: test results, guardrail coverage, and behavioral results, laid out clearly enough that someone outside your head can follow the reasoning.

• At least 4 years building production software, with 1–2 of those years specifically spent building LLM-powered agents — not chatbots, not RAG pipelines, agents that make tool-use decisions on their own.
• Real hands-on time with at least one agent orchestration framework — LangGraph, Bedrock AgentCore, CrewAI, AutoGen, or similar — enough to explain why you'd pick one over another for a given problem.
• Working knowledge of the Model Context Protocol: building or integrating MCP servers and clients, and understanding where tool-selection tends to break down.
• A real prompt engineering practice — versioning changes, testing them against a set of cases before shipping — not just iterating in a playground until it feels right.
• Comfort writing precise specs: boundaries, failure modes, test criteria. This team documents everything and peer-reviews it, so this isn't optional.
• Enough AWS familiarity (Lambda, IAM roles, Bedrock) to understand how your agents actually run, even though infrastructure itself belongs to the Cloud Engineers.

• Experience designing systems where one agent does something and a different agent checks it — separation of duties built into the architecture, not just documented afterward.
• Any exposure to red-teaming or adversarial testing of AI systems, even if it wasn't your main role.
• Time spent in a regulated environment (SOX, SOC2, GDPR, ISO27001), where "it works" was never the whole bar.
• Some familiarity with how agent memory works — short-term session memory versus something more durable — and what each choice costs you in auditability.

#LI-CV1