Platform Engineer (Agent Release)

An agent doesn't go from a developer's laptop to production in one step here — it moves through a staged pipeline with real gates at every stage, then spends its first two weeks in production under close review before anyone trusts it to act on its own. As a Senior Agent Release Engineer, you'll own that entire path: the CI/CD pipeline every agent and every model change moves through, the automated checks that decide whether a release is safe to promote, and the rollback plan for when something isn't. You'll also own model tiering — making sure each agent is running on the model tier that actually fits its task, since that's one of the biggest cost levers on a platform running well over a hundred agents at once. If you like turning "we hope this works" into "here's the automated gate that proves it," this is that role.

Skills and competencies:

• Own and continuously improve the CI/CD pipeline every agent and every model or prompt change moves through, from initial commit through sandbox, development, and staging, before anything reaches production.
• Design and maintain the automated gates that decide whether a release is safe to promote — evaluation score thresholds, security scan results, guardrail test results — so a human isn't manually eyeballing test output before every promotion.
• Own progressive rollout for model and agent changes: canary a change to a small slice of traffic or sessions first, watch the quality and cost metrics live, and only promote to full traffic once those metrics hold.
• Build and maintain the rollback path for every stage of the pipeline, so that a bad release can be reversed in minutes, not hours, and so the trigger conditions for rollback are automatic rather than someone noticing a problem by chance.
• Own model tiering across the agent fleet: routing agents to the model tier that actually fits their task complexity, and adjusting that routing as usage patterns or model pricing change.
• Manage the shadow-period process for newly promoted agents, making sure the right metrics and review checkpoints are actually in place and visible before an agent moves from shadow to fully autonomous operation.
• Run model version regression testing whenever a new model version becomes available, confirming existing agents don't silently degrade before that version gets adopted anywhere in production.
• Partner with the Agent Runtime Engineer on the runtime side of releases to quickly troubleshoot issues in case of a bad rollout. Must-Have:
• At least 5 years in CI/CD, release engineering, or MLOps, including 1–2 years releasing LLM or agent-based systems specifically.
• Has designed and run a progressive rollout strategy (canary, shadow, or blue-green) for something stateful or model-driven, with automatic rollback triggers built in from the start, not a person noticing a problem and rolling back by hand.
• Has gated a release on evaluation scores before, treating a regression as a hard stop the same way a failed test would, and can design that same kind of check to catch a new model version quietly degrading an existing agent before it touches production.
• Strong Python (or similar) and solid infrastructure-as-code skills. Terraform, CDK, GitHub Actions, or CodePipeline all count.
• Comfortable building and deploying containerized services with Docker, and owning their release path through CI/CD end to end. Strong Advantage:
• Direct experience with AWS Bedrock, or something comparable, including its deployment and versioning quirks.
• Has made model tiering or cost-routing decisions — matched task complexity to model cost instead of defaulting everyone to the priciest option.
• Worked somewhere release pipelines needed to produce evidence for an audit, not just pass internally.
• Can explain the difference between traditional CI/CD (pass or fail) and AI release gating (thresholds and probability) to someone who's only worked in the former. That gap trips people up more than it should.

#LI-CV1