AWS lets an AI agent join incident investigations

AWS bundles metrics, logs and traces in CloudWatch Omni. During incidents, an AI agent helps the team look for the cause.

By Alec Chizhik September 26, 2026 3 min read

AWS lets an AI agent join incident investigations

When an application fails, operations teams search through metrics, logs and traces in separate tools while the downtime clock keeps ticking. AWS offers Amazon CloudWatch Omni, an AI-powered observability platform that bundles signals and presents incidents as a shared investigation session. The service has been generally available since 22 September 2026.

Key takeaways

CloudWatch Omni is generally available. AWS runs the service as a standalone web application with single sign-on access, independent of the AWS Management Console.
The DevOps Agent joins incident investigations. It links events across services, traces possible failure paths and suggests next investigation steps.
AI agents get their own tools. Trace Explorer shows model calls hierarchically, and 17 evaluators score responses and tool selection.

Omni runs alongside the AWS console

On 22 September 2026, AWS made Amazon CloudWatch Omni generally available. Omni runs as a standalone web application with its own URL, outside the AWS Management Console. Sign-in uses single sign-on via IAM Identity Center, with support for Okta, Entra ID and SAML 2.0. Team members need no console access. Existing customers enable the service via the “Try CloudWatch Omni” button.

What is Amazon CloudWatch Omni? Amazon CloudWatch Omni is an AI-powered observability platform from AWS for applications and AI agents. It runs as a standalone web application outside the AWS Management Console and uses existing CloudWatch metrics, logs, traces and alarms. In an incident it opens a shared investigation session where the Amazon DevOps Agent links events and suggests next investigation steps.

Omni ingests existing data

Omni detects services and their dependencies automatically from existing data and shows them as a topology. Teams set targets for availability, latency budget and error rate. When the system changes, Omni adjusts the alarms. Omni uses existing CloudWatch metrics, logs, traces and alarms without teams having to rebuild anything. Data from other environments arrives via OpenTelemetry and Omni’s own connectors.

During an incident the team investigates together

If the error rate rises, Omni opens an investigation session with the relevant signals: recent deployments, latency outliers and the service graph. Several team members work together in the same session.

The DevOps Agent joins in

The Amazon DevOps Agent is active by default in investigations. It links events across multiple services, traces possible failure paths in the dependency structure and suggests next investigation steps. Telemetry queries also work in natural language.

AI agents can be monitored too

For generative AI, Trace Explorer shows model calls, tools and processing steps in a hierarchical view. Teams compare different system prompts or model configurations and check which setting an agent needs for its task. Seventeen built-in evaluators score coherence, usefulness, factual accuracy and tool selection. Operations teams monitor AI agents and classic applications in the same tool.

For local development, an IDE extension works without an AWS account. Persistent storage requires a connected AWS account. AWS publishes prices and available regions on the CloudWatch pricing page.

Frequently asked questions

What does Amazon CloudWatch Omni cost?

AWS publishes prices and regions on the CloudWatch pricing page. Existing customers enable the service first via the “Try CloudWatch Omni” button.

Do team members need access to the AWS Management Console?

No. Omni is a standalone web application with its own URL. Sign-in uses single sign-on via IAM Identity Center, with support for Okta, Entra ID and SAML 2.0.

Can Omni be used for environments outside AWS?

Yes. Alongside existing CloudWatch data there is an OTLP endpoint for OpenTelemetry and connectors for environments outside AWS. For local development, an IDE extension is enough, with no AWS account required.