Amazon Web Services made its DevOps Agent generally available on March 31, 2026, according to AWS. The product is real: a multi-agent system that investigates production incidents, correlates telemetry across your stack, and recommends fixes. It cannot execute any of those fixes — no code pushes, no rollbacks, no security group changes — but it will tell you what to do. The architecture is solid. The pricing is a retention play.
Starting April 10, AWS DevOps Agent costs $0.0083 per agent-second. That number is almost beside the point. AWS Support customers receive monthly credits equal to a percentage of their prior month's support bill: 100 percent for Unified Operations, 75 percent for Enterprise Support, and 30 percent for Business+ plans. A company spending $10,000 per month on Enterprise Support gets $7,500 in DevOps Agent credits — effectively halving the cost of a product its largest customers were already paying to access through their existing support contracts. The companies spending the most on AWS are the ones who get it cheapest. This is not an incident response product. It is a loyalty program with a demo attached.
The multi-agent architecture is worth understanding on its own terms. A lead agent acts as incident commander: it parses the symptom, builds an investigation plan, and delegates to specialized sub-agents that each handle a slice of the problem. Those sub-agents operate with compressed context windows — a deliberate trade-off that keeps individual agents fast and focused rather than trying to reason across the entire incident at once. The lead agent synthesizes their findings. AWS has published the pattern, including five mechanisms the team identified as necessary to move from prototype to production: evaluations, visualization, fast feedback loops, intentional change management, and reading production samples. That is honest engineering writing, and it suggests the team actually shipped something.
Preview customers report meaningful numbers. Western Governors University reduced resolution time from an estimated two hours to 28 minutes during a service disruption scenario. Commonwealth Bank of Australia found root causes in under 15 minutes that veteran engineers estimated would take hours. Zenchef, a restaurant management software company, used the agent during a hackathon to identify an IAM misconfiguration in under 30 minutes. These are self-reported case studies from AWS's marketing channels — treat them as signals, not benchmarks. But the WGU and Commonwealth Bank data points are directionally credible because the scenarios are specific and the outcomes are concrete.
The agent cannot do anything without a human in the loop. It cannot push code. It cannot roll back deployments. It cannot restart services. It cannot modify security groups. It investigates and recommends. AWS is explicit about this — the product page and the testing coverage both state it plainly — but the framing of "autonomous incident response" in the announcement may set different expectations. For teams that want an agent to actually execute a fix, this is a critical constraint. For teams that want a very fast junior engineer reading logs and making suggestions, this is the product.
The support credit structure is the detail that changes how you read the rest of the announcement. AWS has a support tier problem at the high end: large customers pay enormous monthly fees and still have to staff their own on-call rotations. The DevOps Agent is not going to replace that staffing. But it is going to make the support contract feel more valuable to the people paying $10,000 or more per month — which is exactly the segment AWS most wants to retain. Swami Sivasubramanian, AWS's vice president for agentic AI, served on the National AI Advisory Committee from May 2022 through May 2025, giving the product a proximity to policy conversations that is probably not coincidental.
Pricing examples AWS publishes illustrate the math: a small team running 10 investigations per month at 8 minutes each pays $39.84. An enterprise running 500 incidents plus 40 evaluation runs per month pays $2,290.80. These are not large numbers relative to a typical AWS bill at the Enterprise Support tier. The credits cover most or all of the cost for any customer where the support contract is already a significant line item.
The agent integrates with CloudWatch, Datadog, Dynatrace, New Relic, Splunk, Grafana, GitHub, GitLab, ServiceNow, PagerDuty, and Slack. It also supports Azure and on-premises environments via the Model Context Protocol (MCP), an open standard that AWS endorsing is more significant than the agent itself — MCP adoption by a hyperscaler validates the protocol in a way that community adoption alone had not. The agent launched in US East, US West, Europe (Frankfurt and Ireland), and Asia Pacific (Sydney and Tokyo) regions.
The honest read: AWS DevOps Agent is a capable, honestly-designed incident investigation tool wrapped in a pricing structure designed to reduce churn among the customers AWS can least afford to lose. If you are already spending $5,000 per month on AWS support, the credits make this effectively free. If you are not, $0.0083 per agent-second is reasonable for what it does — as long as you understand what it does, which is investigate, not act. The gap between those two things is where incidents still live.
