Deriv achieves 90% first-pass root-cause accuracy with AWS DevOps Agent

Over more than two decades, Deriv has continually modernized the technology supporting its online trading services, moving from a legacy environment toward a distributed architecture spanning multiple platforms and services. This transformation increased the complexity of managing production operations for services that run around the clock.

When incidents occurred, engineers manually reviewed logs, checked services and infrastructure, and traced dependencies across systems to identify the root cause. An issue could surface in one part of the environment but originate from an upstream or downstream service. Depending on the complexity, resolving an incident could take four to six hours, followed by several more hours of post-incident analysis.

As Deriv expanded its use of AI across software development and operations, the company saw an opportunity to apply agentic AI to production operations. Deriv wanted to accelerate incident investigation and build a clearer understanding of dependencies and potential risks across its infrastructure.

Najib Huq, engineering senior manager at Deriv, says “We are constantly evolving our infrastructure and releasing new services, so we want our teams focused on improvement rather than dealing with outages.”

To reduce the manual effort involved in incident investigation, Deriv began using AWS DevOps Agent, an agentic AI solution that investigates operational issues by analyzing logs, applications, infrastructure, and dependencies across the environment. The agent draws context from across Deriv’s environment, including Amazon Elastic Compute Cloud (Amazon EC2), Amazon Simple Storage Service (Amazon S3), Amazon Elastic Block Store (Amazon EBS) and Amazon EventBridge. It also spans Kubernetes environments, hosted databases, and multiple AWS accounts. Deriv extends this context with additional documentation, skills, and custom connectors for resources outside AWS.

“What makes AWS DevOps Agent different is the context it carries across our environment. It can connect changes and dependencies that would otherwise take engineers much longer to piece together,” says Huq.

Deriv integrated AWS DevOps Agent into its incident workflow so monitoring alerts automatically trigger investigations through a webhook. When issues surface through other channels, such as customer reports, engineers initiate investigations from the AWS DevOps Agent console, where they follow progress and review identified root causes.

Beyond incident investigation, Deriv uses AWS DevOps Agent for release risk reviews and proactive health checks to identify potential issues before updates reach production. For some low-code services, Deriv exports relevant configuration and service information to GitHub, where the agent analyzes it alongside repository data when reviewing proposed changes. In one pull request, AWS DevOps Agent identified that a proposed change would fail because of a configuration dependency in another repository—an issue Deriv’s existing security and code-quality agents had not detected. “The value is being able to look beyond where an issue appears and connect it to what changed elsewhere in the environment. That gets us closer to the real cause much faster,” says Huq.

By combining automated incident investigation with contextual analysis across its environment, Deriv has achieved 90 percent first-pass accuracy in autonomous root-cause analysis. This reduces the need for engineers to validate multiple hypotheses or repeatedly redirect an investigation. With previous AI-based approaches, checking an incorrect hypothesis could add another 20–30 minutes of investigation.

“First-pass accuracy is what gives AWS DevOps Agent the edge. Every incorrect hypothesis can send an engineer into another investigation before getting to the real problem,” states Huq.

Deriv has also reduced MTTR by at least 50 percent, shortening incidents that previously required hours of investigation. Faster resolution reduces the time engineers and senior leaders spend responding to outages and gives technical teams more time for infrastructure optimization, AI initiatives, new applications, and service improvements.

Beyond resolving incidents faster, Deriv is shifting its focus toward preventing incidents. Proactive health checks help the team identify emerging issues and address them before they develop into outages, while release-risk reviews flag potentially disruptive changes before deployment.

As Deriv expands its use of AI-generated code and infrastructure changes, the company plans to extend AWS DevOps Agent across more accounts and repositories, with the goal of preventing high-risk releases from reaching production. Deriv is also exploring how the agent can identify opportunities to optimize infrastructure costs by consolidating or removing unnecessary resources.

“The biggest opportunity for us is prevention. We want to identify risks before they become incidents, so our teams can spend more time building and improving rather than responding to outages,” says Huq.