AWS launches CloudWatch Omni, AI-powered observability for apps and agentic workloads

CloudWatch Omni reimagines AWS's flagship observability service for the agent era. Traditional monitoring assumes deterministic services; Omni instead treats AI agents and applications as first-class citizens in one unified experience, with auto-discovered topology maps, natural-language querying, and AI-guided root-cause investigation driven by the AWS DevOps Agent.

The agentic-specific half is the differentiator. Omni delivers observability purpose-built for generative AI and agentic workloads, letting developers trace, evaluate, and experiment with agents across any framework — from an IDE or a standalone web experience — using open standards and built-in evaluators that score quality, correctness, and coherence. That directly addresses a growing pain point: as teams ship agents, they lack the equivalent of APM tooling to understand why an agent chose a tool, followed an instruction, or hallucinated.

Mechanically, it complements the rest of AWS's agent stack shipped this week — AgentCore Runtime V2, Strands Evals, and the Strands harness — forming what Forkast called a now-complete 'cloud agent infrastructure layer.' The pitch is that AWS owns the full loop: build (Bedrock/AgentCore), evaluate (Strands Evals), run (AgentCore Runtime), and observe (CloudWatch Omni).

Competitively, this pushes AWS ahead of standalone observability vendors (Datadog, Grafana) and cloud rivals on agent-native monitoring, though those incumbents are racing to add similar evaluators. The caveat is lock-in: Omni's deepest value comes when the whole agent stack lives on AWS, and 'any framework' support will be judged by how well it handles non-AWS agents in practice. What to watch: adoption among teams already running agents in production and whether the built-in evaluators are trusted enough to gate deployments.