AWS made AWS DevOps Agent generally available on 31 March 2026. It investigates production incidents on its own, across AWS, Azure, and on-premises environments, and works alongside your team through the full incident lifecycle.
AWS reports that customers and partners using the agent during preview saw up to 75 percent lower MTTR, 80 percent faster investigations, and 94 percent root cause accuracy.
This post explains what it does, how it differs from how most teams handle incidents today, what the GA release added, and what it costs.
The problem it addresses
AWS describes the situation directly. When an incident hits, on-call engineers are under pressure to find the cause while keeping stakeholders informed. They analyze data across several monitoring tools, review recent deployments, and coordinate the response. Once service is restored, teams rarely have the bandwidth to turn what they learned into lasting improvements.
Most of that time is not spent thinking. It is spent gathering context, and gathering context is a task a machine can do faster.
How incident response changes
Today, in most teams. An alert fires. An engineer opens the metrics dashboard, then the logs, then the deployment history, then the traces. They form a hypothesis, check it, and repeat. Each tool is a separate window with a separate mental model. The clock runs while the engineer assembles a picture.
With the agent. The investigation starts the moment the alert arrives, at 2am or during a peak. The agent already knows your applications and how they depend on each other. It pulls telemetry, code, and deployment data together, correlates them, and presents a likely root cause with the evidence behind it. The engineer arrives to a hypothesis instead of a blank screen.
The difference is not that the agent is smarter than your engineers. It is that it reads every source at once, has no context to reload, and does not get tired at 2am.
What the agent does
The agent covers three kinds of work.
- Investigations. It responds to an incident, correlates signals across your tools, and guides the team to resolution. AWS positions it as guiding teams rather than executing changes itself.
- Evaluations. It examines your systems for reliability risks before they trigger an alert, and analyzes patterns across past incidents to recommend changes that prevent repeats.
- On-demand SRE tasks. A conversational assistant for questions about your architecture, metrics, alarm status, deployment history, and incident patterns. It can produce custom charts and reports.
It also learns. The agent builds skills from how your team resolves particular incident types, and you can add custom skills that encode your own procedures and institutional knowledge.
What it connects to
You connect the agent to your existing tools through the AWS Management Console.
- Telemetry. Amazon CloudWatch, Datadog, Dynatrace, New Relic, Splunk.
- Deployments. GitHub Actions, GitLab CI/CD.
- Incidents. ServiceNow built in, PagerDuty and others through configurable webhooks.
- Everything else. Bring-your-own MCP servers for custom internal tools and open source stacks such as Grafana and Prometheus.
An Agent Space defines what the agent can reach, using IAM roles to introspect resources across your accounts. You can scope a space to one application, to an on-call team covering several services, or run a central space.
What GA added over the preview
If you looked at the preview and set it aside, several things have changed.
- Azure support. The agent now investigates Azure workloads and correlates across multicloud deployments.
- On-premises support. Through MCP, it discovers on-premises resources by analyzing metrics, logs, and code, and builds a topology from them.
- Triage. It assesses incident severity automatically and detects duplicate tickets, linking them to the primary investigation to cut noise.
- Custom and learned skills. You can encode your own investigation procedures, and the agent develops its own from your team's patterns.
- Charts and reports. Ad hoc analysis you can save and share.
What it costs
Billing is per second of active agent time. Nothing is charged while the agent is idle, so cost tracks the volume and length of investigations rather than a subscription.
New customers get a two-month free trial, and AWS DevOps Agent is included in the AWS Free Tier for new AWS accounts. AWS Support customers also receive monthly credits based on the prior month's support charge: 100 percent for Unified Operations, 75 percent for Enterprise Support, and 30 percent for Business Support+. For many teams that offsets most or all of the cost, so check your support tier before estimating spend.
What determines the results you get
The agent reasons over what you connect to it. In practice, the range in reported outcomes comes down to the state of the environment underneath.
Correlated telemetry matters most. If metrics, logs, and traces live in separate tools with no shared identifier, there is less to correlate. The same applies to deployment data that is not linked to the services it changes, and to dependencies that exist in people's heads rather than in any system the agent can read.
This is worth assessing honestly before deployment. It is also ordinary engineering, and it pays off whether or not you adopt an agent.
At NileForge, this is the work we do with teams on AWS: getting telemetry, topology, and deployment data into a state where tooling like this performs at the top of its range rather than the bottom.
If you are evaluating AWS DevOps Agent for your environment, talk to our team.