Skills and competencies:

• Own the health of the agent execution environment day to day — session isolation, resource limits, and catching the failure modes that only appear once real concurrency and real traffic patterns show up, not just in a clean test run.
• Investigate and resolve cases where an agent behaves differently in production than it did in testing, working directly with the Agent Engineer or Quality Engineer who built it to figure out whether the problem is the agent's design or the environment it's running in.
• Tune cold-start and concurrency settings for the platform's critical-path functions, and review them on a regular cadence as usage patterns shift rather than setting them once and forgetting them.
• Understand the platform's session isolation model well enough to reason about its limits — where isolation is guaranteed by the underlying compute layer, and where the platform has to add its own controls because that guarantee doesn't fully hold.
• Build and maintain the observability that lets someone answer "is this agent actually working correctly," not just "is it technically up" — tracing, behavioral drift detection, and quality signals sitting alongside the usual metrics and logs.
• Track per-agent and per-session cost and efficiency, and flag agents that are burning more tokens, calling more tools, or running longer than the task should reasonably require.
• Deploy and manage runtime-layer infrastructure resources through existing CI/CD pipelines as needed — provisioned concurrency settings, runtime-specific IAM roles, observability configurations, etc.
• Run and improve the tests that validate isolation actually holds — for example, confirming that one agent's session genuinely can't reach or affect another's, not just assuming it because the platform is supposed to guarantee it.

• At least 5 years in a platform, SRE, or infrastructure engineering role, with real production experience running serverless or containerized workloads on AWS (Lambda, Fargate, Docker, Kubernetes or equivalent) at meaningful scale.
• Hands-on experience with AWS observability tooling (CloudWatch, X-Ray, or a comparable distributed tracing stack) — able to go from "something's wrong" to a root cause using traces and logs, not just dashboards.
• Real experience deploying and managing infrastructure through CI/CD independently — comfortable owning IaC changes (Terraform, CDK, or similar) end to end rather than handing them off to someone else.
• Working understanding of compute isolation concepts (containers, microVMs, or similar sandboxing models) and where their guarantees actually stop, since a lot of this role is reasoning about the edges of what a platform promises versus what it might not fully cover on its own.
• Comfort investigating cost and performance problems at a granular level — able to trace an unexpectedly expensive or slow workload back to a specific cause, not just flag that costs went up.
• Strong incident response instincts: staying calm and methodical while root-causing a live production issue, and following through with an actual fix rather than a workaround.

• Direct experience with AWS Bedrock, SageMaker, or any managed AI/agent runtime platform.
• Experience with cold-start or concurrency tuning specifically (Lambda provisioned concurrency, container warm pools, or similar).
• Exposure to a security-conscious or regulated environment where infrastructure changes go through a formal review or approval process.
• Familiarity with LLM-specific cost drivers (token pricing, tool-call volume, model tiering) even if it wasn't the primary focus of a past role.

#LI-CV1