AWS DevOps Agent: Worth It?

AWS DevOps Agent is now GA — and it’s in Sydney!

At AWS re:Invent 2025, AWS introduced DevOps Agent as a service designed to be “your always-available operations teammate that resolves and proactively prevents incidents, optimizes application reliability and performance, and handles on-demand SRE tasks.” With general availability, AWS has expanded beyond the initial preview, adding support for more real-world scenarios including hybrid, on-premises, and multi-cloud environments.

Once enabled, DevOps Agent begins by learning your environment: your resources, configurations, and telemetry. From there, it can detect issues, investigate what’s happening, and suggest likely root causes. By the time a human operator steps in, a large portion of the initial triage may already be done.

That sounds pretty awesome… but also a little bit scary.

So, what’s it actually like to use in practice — and what does it cost?

What is AWS DevOps Agent?

AWS DevOps Agent isn’t just a simple AI tool. It’s what AWS refers to as a frontier agent: an autonomous system designed to operate independently, scale as needed, and run with minimal human oversight.

It is designed to help teams investigate incidents, identify root causes, and prevent recurrence by analysing signals across your environment. To do this, it pulls together data from services like Amazon CloudWatch, AWS Systems Manager, and resource APIs to build a picture of what’s happening when something goes wrong.

According to AWS, DevOps Agent works by:

• Learning your resources and their relationships
• Integrating with your observability tools, code repositories, and CI/CD pipelines
• Correlating telemetry, code, and deployment data to understand how your application behaves
• Supporting applications across multi-cloud and hybrid environments

Core capabilities

There are three key areas that DevOps Agent focuses on:

Incident investigation

DevOps Agent can take a natural language prompt and begin investigating issues immediately. Instead of manually jumping between metrics, logs, and dashboards, it gathers relevant data and presents a summary of what it believes is happening.

In practice, this means pulling:

• Metrics from CloudWatch
• Logs from CloudWatch Logs
• Recent changes or events
• Resource configuration details

All of this gets stitched together into a single story.

Root cause analysis

Once it has the data, the agent attempts to correlate signals and identify likely root causes. This is where things get interesting — it’s effectively doing the “join the dots” work that operators normally do manually.

Rather than just showing symptoms, it aims to answer a much more useful question:

What actually caused this?

Recommendations and remediation guidance

Finally, DevOps Agent suggests possible next steps. These might include configuration changes, scaling actions, or further investigation paths.

It’s important to note that, at least in its current form, this is primarily advisory. It helps you understand and decide. DevOps Agent isn’t making changes for you. A human still needs to do that side.

What’s it actually like?

My colleague Tony and his customer both attended AWS re:Invent 2025 and drank the DevOps Agent cool-aid. While I’ve only played with it in my own environment, he was able to put the preview through its paces in a real-world setting.

When you configure DevOps Agent, you start with an Agent Space. This defines the boundary of what the agent can see and interact with, and it’s not restricted to a single account within your Organisation.

Tony tested this in a development environment that spanned multiple accounts and included hundreds of services, Lambda functions, and supporting resources. Quite impressively, the agent was able to map out the services and their interconnectivity across the environment. With that many components, this process took some time, and the resulting view was understandably complex.

AWS recommends using Agent Spaces to represent smaller, application-focused environments. In practice, that makes a lot of sense. Smaller, more targeted spaces provide clearer visibility and allow the agent to focus more effectively. Or more accurately, they make it easier for the human operator to interpret what’s going on.

Tony has pointed out that having too many Agent Spaces isn’t ideal, either. If you have similarly architected environments, grouping them into a single Agent Space may make management easier.

One thing that stood out was the agent’s ability to infer workload structure based on naming conventions. Tony’s team follows consistent, well-structured naming practices for Lambda functions and other resources, which helped the agent quickly establish relationships and understand how components were expected to interact.

It’s a good reminder of the old rule: garbage in, garbage out. The better your environment is structured, the better the results you’ll get from tools like this.

While DevOps Agent can integrate with tools like Slack and ServiceNow, it currently doesn’t make changes within your AWS environment. It operates in a read-only, advisory capacity.

In a world where AI is still finding its place in production environments, that’s a reassuring design choice. The agent can investigate, analyse, and recommend — but the final decision, and the execution, still sits firmly with a human.

What does it cost?

For you? I’ll give you a couple of months free! (Well, AWS will.)

At the time of writing, AWS provides a free tier for the first two months:

“Each trial month includes up to 10 agent spaces, 20 hours of investigations (incident response), 15 hours of evaluations (incident prevention), and 20 hours of on-demand SRE tasks (chat). Standard pay-as-you-go pricing applies for usage beyond these limits and after your trial ends.”

The included amounts give you a great opportunity to try out AWS DevOps Agent. You can then review what your spend would have been and decide if it works for you.

One of the more interesting aspects — and something I haven’t really seen before — is a discount for those on AWS Support contracts. The What’s New post states:

AWS Support customers receive monthly DevOps Agent credits based on the prior month’s gross AWS Support spend: 100% for Unified Operations, 75% for Enterprise Support, or 30% for Business Support+.

Outside of the free tier or support discounts, the costs for DevOps Agent are:

• $0.0083 per agent-second for investigations (incident response)
• $0.0083 per agent-second for evaluations (incident prevention)
• $0.0083 per agent-second for on-demand SRE tasks (chat)

This “agent-second” model reflects the time the agent spends analysing, correlating, and generating responses. It’s a slightly different way of thinking about cost compared to typical request-based pricing.

You’ll also want to consider any underlying costs from services like Amazon CloudWatch, especially if large volumes of logs or metrics are queried as part of the investigation.

What’s next?

What now? Well, you’ve got two free months, so give it a go!

I’d recommend keeping your Agent Space targeted. Focus on a single application, ideally one that’s well structured. When you’re trialling the service, you want to see what it can do at its best.

Then try a second Agent Space with something a bit messier. I don’t want you to be discouraged by results that aren’t great due to the environment. It’s sort of like doing a Well Architected review of your entire environment rather than the workload it’s designed for. You’ll likely get something useful, but not optimal.

This service doesn’t replace your ops team — it enhances them. It reduces the time spent gathering and correlating data, but the judgment, context, and decision-making still sit with your engineer.