Deployment releases an AI system or part of it into an environment serving users or production workloads. A release can change context sources, guardrails, or orchestration—not just application code. Responsibilities vary by adoption model, from vendor configuration to model serving and infrastructure. section 9.3.8
Two rollout strategies reduce release risk:
- Blue-green deployment keeps current and new versions in equivalent environments, then switches traffic after validating the new version; the previous environment enables rapid rollback. section 9.3.8
- Canary release exposes the new version to a small share of users or requests before expanding rollout, allowing real-world monitoring. section 9.3.8
Because small context changes can alter behavior, releases should be tied to evaluation, observability, rollback, and governance. Monitor quality, latency, cost, safety, and task completion during rollout. section 9.3.8




























