AWS CloudWatch Omni goes after the hardest question in agentic AI: Why did the agent do that?
SiliconANGLE Zeus Kerravala
AWS just launched CloudWatch Omni to explain why an AI agent made a move, not just whether it ran. That matters because agents can look healthy and still be wrong.
Based on reporting by SiliconANGLE, Zeus Kerravala — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Amazon Web Services has pushed CloudWatch into a new job: figuring out why an agent did something, not just whether the system stayed up. CloudWatch Omni, which became generally available last week, is AWS’s answer to the messier world of agentic AI, where an agent can hit its latency target, avoid errors and still give the wrong answer or choose the wrong tool.
That’s the problem traditional observability misses. A clean dashboard does not tell you whether the model pulled stale knowledge, routed a request badly or followed a broken path through a workflow. Omni is AWS’s attempt to move observability from infrastructure health to decision quality, and that’s a much harder business question. AWS says IDC expects more than 1 billion deployed agents by 2029, which gives you a sense of the scale of the headache.
The center of the product is evaluation. Omni ships with 17 built-in evaluators that score things like coherence, helpfulness, faithfulness and routing correctness. Teams can compare prompt versions, build test datasets from production traffic and catch regressions automatically. Those evaluators can also run continuously on live traffic, so quality drift shows up before customers do. Sony is already using it, and the company’s enterprise agentic AI platform now supports hundreds of proof-of-concept and production workloads.
AWS is also trying to pull the work out of the console and closer to where developers actually build. Omni has native extensions for Visual Studio Code, Cursor and Kiro, shows traces locally without requiring an AWS account, and shares a data layer with the web experience used by operators. That matters because the same trace should be visible to the engineer fixing it and the operator investigating it. It also means quality work happens earlier, when it is cheaper.
The other big move is the unified data store. Agent traces, application telemetry and infrastructure signals all sit together in CloudWatch, so an investigation can move from a bad tool result to an API error to an exhausted database connection pool without jumping between systems. Capital One, a design partner, said it wanted topology-aware intelligence and natural-language querying across telemetry with full data ownership through OpenTelemetry. That ownership piece is not cosmetic in banking and other regulated sectors; audit trails matter.
AWS is betting on openness, but not too much. Omni supports OpenInference and the AWS Distro for OpenTelemetry, and it works with tools and frameworks including LangChain, LangGraph, CrewAI, the OpenAI Agents SDK, Strands and the Vercel AI SDK. It also supports Azure ingestion today, with more multicloud coverage promised later. But the intelligence layer still lives on AWS, which is exactly how AWS likes these stories to end.
My take — AI-written commentary, not fact-checked reporting
This is AWS admitting that “is it up?” is a childish question for agentic AI. The real product now is judgment, and whoever controls the evaluator controls the story. OpenTelemetry keeps the doors open; the clever bit is making sure the keys still sit on AWS.
Read more about this at: SiliconANGLE