Amazon CloudWatch Omni Puts Agents and Apps in One AWS Telemetry View
Amazon CloudWatch Omni reached general availability on 22 September 2026, unifying AI agent and application observability with 17 evaluators.
Amazon CloudWatch Omni is now generally available, giving AWS customers one interface for tracing AI agents and conventional applications side by side. AWS released the service on 22 September 2026 as an extension of CloudWatch rather than a replacement, and it arrives in two forms: a standalone web application with enterprise single sign-on, and a free VS Code extension. The pitch targets a specific operational gap, because teams that instrument agents in one tool, applications in another and infrastructure in a third have no shared timeline when a production incident starts.
That gap is measurable in workflow terms. Diagnosing an agent's behaviour previously meant moving between Amazon Bedrock AgentCore for agent traces and evaluations, application performance monitoring for services, and CloudWatch for infrastructure metrics. Omni collapses those into a single view without forcing a migration, since existing CloudWatch alarms, dashboards, APIs and console workflows keep running unchanged.
What Amazon CloudWatch Omni Actually Ships
The service supports agent frameworks including LangChain, LangGraph, CrewAI, the OpenAI SDK and the Vercel AI SDK. Its foundation rests on open standards: OpenTelemetry, OpenInference and the AWS Distro for OpenTelemetry, known as ADOT. Framework breadth matters because a team can keep its existing agent code and redirect only the telemetry.
Two capabilities carry most of the product. A Trace Explorer renders hierarchical timelines of LLM calls and tool invocations, so an investigation can run from a user request down to the individual tool call that misfired. A set of 17 built-in evaluators scores output on coherence, helpfulness and tool-selection correctness, which converts agent quality from a review-time judgement into a tracked metric.
The IDE extension carries a second implication. Because agents can be traced, evaluated and experimented with from inside the editor, quality checks move into the development loop instead of waiting for production telemetry. A developer can compare two prompts or tool configurations against the built-in evaluators before a change ships, which is the same shift unit tests and continuous integration brought to application code.
| Capability | Detail |
|---|---|
| Built-in evaluators | 17, covering coherence, helpfulness, tool-selection correctness |
| Framework support | LangChain, LangGraph, CrewAI, OpenAI SDK, Vercel AI SDK |
| Open standards | OpenTelemetry, OpenInference, ADOT |
| Access points | Web experience with enterprise SSO; free VS Code extension |
| Migration impact | Existing CloudWatch alarms, dashboards, APIs unchanged |
Teams already running on Bedrock AgentCore receive coverage automatically. AgentCore's evaluators for retrieval quality, tool selection, guardrail enforcement and memory coherence appear in Amazon CloudWatch Omni out of the box, and existing AgentCore Observability telemetry lands there without re-instrumentation. AWS added auto-discovered topology, natural-language queries and AI-guided investigation driven by the AWS DevOps Agent.
The off-console delivery changes who can use the tool as much as how. CloudWatch's traditional audience is DevOps engineers, developers, site reliability engineers, IT managers and product owners, and console access has usually been gated to a narrower group. A web experience with enterprise single sign-on lets product and support staff read the same traces their engineers use, which shortens the path from a customer complaint to the specific agent decision behind it. AWS states that general availability covers a limited set of regions at launch rather than the full global footprint.
The Competing Approaches
Buying into Omni is one of three credible paths, and the choice is less obvious than the launch framing suggests.
Single-platform consolidation is the simplest. Instrument once, accept AWS as the system of record for traces, evaluations and dashboards, and gain one timeline that spans agents, services and infrastructure. The cost is concentration: history, alerting and evaluation logic accumulate inside one vendor's tooling.
Open-standards instrumentation is the hedge. Because Omni ingests OpenTelemetry and OpenInference, a team can instrument agents once and keep the option of exporting the same spans to a competing backend later. AWS has leaned on that interoperability argument explicitly. The hedge is partial, since the evaluation layer and investigation workflows stay AWS-native even when raw traces remain portable.
Staying with third-party APM suites is the third path. Many enterprises already license dedicated monitoring platforms for service telemetry and have years of dashboards, runbooks and on-call practice built on them. Omni's arrival does not invalidate those contracts, and for teams whose agents are a small share of total traffic, the consolidation benefit may not clear the switching cost.
The multi-cloud caveat deserves attention. AWS has documented how agents running on-premises, on GCP, on Azure or on developer machines route session traces, span metrics and token usage into its observability services using ADOT and IAM credentials. That path works, and it also confirms the architecture: external agents send their telemetry into AWS services, and the customer manages the associated IAM configuration. Omni widens who can reach the data, while the data itself still terminates in AWS.
Where the Trade-offs Land
Consolidation against portability is the first tension. OpenTelemetry reduces instrumentation lock-in, yet the evaluators, trace history and investigation workflows that make Omni useful are AWS-specific. A team that adopts the full experience is buying depth in exchange for some exit cost.
Coverage against domain depth is the second. Seventeen general evaluators cover coherence, helpfulness and tool selection, which suits teams standardising agent quality for the first time. Regulated or specialised workloads typically need evaluators tuned to their own correctness definitions, and those still have to be written and maintained.
Duplication is a related risk for teams that already built scoring harnesses of their own. AgentCore shipped evaluators for retrieval quality, tool selection, guardrail enforcement and memory coherence, and Omni layers 17 general metrics on top. Organizations now have to decide which layer is authoritative, and keeping two sets of quality metrics alive consumes engineering time even when both are inexpensive to run.
Cost transparency is the third open question. AWS has confirmed that the VS Code extension is free to use, and has not detailed pricing for the wider service. Trace retention, evaluator execution and the volume of spans generated by multi-step agents all scale with usage, so the bill depends on instrumentation decisions teams make long before the invoice arrives.
Governance is the fourth. Off-console access with enterprise SSO widens who can investigate production incidents, which is a genuine operational gain and a permissioning problem at the same time. Data residency is the related constraint, since availability begins in a limited set of regions rather than everywhere CloudWatch runs.
Who Should Move, and When
Teams already standardised on CloudWatch and Bedrock AgentCore have the clearest case. Their telemetry already flows into the service, the migration cost is close to zero, and AgentCore's evaluators arrive without extra work. Multi-framework shops running agents built on LangChain, CrewAI or the OpenAI SDK get a second clear benefit, because Omni removes the need to reconcile several tracing tools.
Organizations with heavy investments in a competing monitoring platform, or with workloads pinned to regions where Omni is not yet available, have reason to wait. The instrumentation work required to adopt Amazon CloudWatch Omni is portable, so a staged approach costs little: adopt the open standards now, keep the evaluation layer optional, and revisit once regional coverage and pricing are public.
Why this matters
Observability is where AI agents stop being demos. Once an agent touches customer data or a payment path, the ability to reconstruct why it acted is an operational requirement rather than a nice-to-have, and AWS has made that capability part of its default monitoring surface instead of a separate product. The strategic signal is that the largest cloud provider now treats agent telemetry as standard infrastructure, which raises the floor for every vendor selling agent evaluation and pushes differentiation toward domain-specific correctness rather than basic trace capture.
Sources
Introducing Amazon CloudWatch Omni: AI-powered observability for generative AI and agentic workloads
Amazon CloudWatch Omni: AI-first observability for agents and applications
Amazon Monitoring and Observability-Amazon CloudWatch - AWS
Monitor on-premises and multi-cloud AI agents with AgentCore Observability | Artificial Intelligence
Related Articles
- AWS Agent Registry Launches to Centralize AI Governance
- AWS and Databricks Launch Agentic Stack to Streamline Enterprise AI Agents
- CloudWatch Coding Agent Insights: Dashboard for AI Developer ROI
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.