Amazon Web Services has moved two of its most significant AI agents into general availability: AWS DevOps Agent and AWS Security Agent. The GA announcement, confirmed in the AWS Weekly Roundup on April 6, 2026, marks a shift from the lab to real enterprise production — and the results from preview customers are hard to ignore.

AWS is calling these “frontier agents,” a term they are using deliberately to separate them from the AI copilots and assistants the market has grown used to. Where a copilot responds to prompts, a frontier agent works autonomously toward a goal — running persistently for hours or days without constant human intervention, scaling massively across concurrent tasks, and handling the full cycle of investigation, remediation, and prevention rather than surfacing suggestions for a human to act on.

What These Agents Actually Do

AWS DevOps Agent takes on site reliability engineering work. It investigates production incidents autonomously by correlating telemetry, code, and deployment data across your entire stack — whether your applications live on AWS, Azure, a hybrid environment, or on-premises. It works with your existing observability tools, runbooks, and CI/CD pipelines rather than requiring a rip-and-replace approach. Once it understands the environment, it can proactively surface issues before they become outages.

AWS Security Agent brings on-demand penetration testing and continuous security review to development workflows. It audits design documents, scans pull requests for vulnerabilities, and enforces an organisation’s security policies automatically — surfacing the risks that matter rather than generating noise.

The two agents address the most expensive pain points in operating modern software at scale: unexpected incidents and security vulnerabilities, both of which have historically required expensive, hard-to-hire specialists.

The Numbers from Preview Customers

The results AWS is citing from preview customers are significant:

• 75% lower MTTR (mean time to resolution) across incidents
• 80% faster investigations
• 94% root cause accuracy
• 3 to 5 times faster overall incident resolution

The most concrete case study comes from Western Governors University. Their SRE team used DevOps Agent to analyse a service disruption scenario. Resolution time dropped from an estimated two hours to 28 minutes — a 77% improvement. United Airlines and T-Mobile are also named as preview customers.

These are not benchmark results. They reflect the agents working on real production environments with the complexity and ambiguity that entails.

The Pricing Model Changes the Calculation

AWS has priced these agents on a consumption basis — roughly $30 per hour for SRE work (DevOps Agent) and approximately $50 per hour for penetration testing (Security Agent). Customers on AWS Enterprise Support and above receive monthly credits based on their support spend, which meaningfully offsets costs for large users.

That pricing model is worth thinking about carefully. A senior SRE engineer in a major city costs $150,000 to $250,000 per year in salary alone, before benefits and management overhead. AWS DevOps Agent running at $30 per hour is materially cheaper for incident investigation work — and it does not need to sleep, take PTO, or hand off context between shifts.

This is not an argument that DevOps Agent replaces engineers. The preview customers who saw the biggest gains used it to augment their SRE teams, not replace them. But for organisations where expert DevOps capacity is a bottleneck — and for the vast majority of mid-sized businesses, it is — the economics are genuinely different from what they were six months ago.

Why This Matters Beyond the AWS Ecosystem

The GA release of these agents is a data point in a broader pattern that every business building on cloud infrastructure should understand.

The definition of “AI agent” is being actively upgraded across the major platforms. What AWS calls a frontier agent — autonomous, multi-hour operation, goal-directed rather than prompt-directed — is what Anthropic calls an agentic workflow, what Microsoft is building toward with Agent 365, and what Salesforce is deploying through its Agentforce platform. The terminology is different, but the direction is consistent.

The specific capability being demonstrated here — AI that can run a full incident investigation across a hybrid environment, correlate data from observability tools, code repos, and runbooks, and resolve the issue without a human in the loop — was essentially impossible 18 months ago. It is now a generally available product with published pricing.

What This Means for Business

For engineering and operations teams, the practical question is straightforward: where is incident response currently your biggest bottleneck, and how much of that work is investigation versus judgement?

AWS DevOps Agent is strongest at the investigation and correlation phase — the part that is time-consuming but largely mechanical. It connects telemetry to deployment events to code changes and surfaces a root cause. What it does less well, by design, is make the human judgement calls: whether to roll back versus patch forward, how to communicate with customers, what the risk appetite is for a particular fix. Those decisions stay with your team.

For security, the Security Agent’s penetration testing capability is most valuable in organisations where security reviews are happening too infrequently because qualified reviewers are scarce. Automating the baseline scan and policy check frees security engineers for the work that actually requires expertise.

The broader signal is that agentic AI has cleared a threshold. It is no longer a research concept or a heavily caveated preview. It is a production workload running on enterprise infrastructure at companies like United Airlines and T-Mobile, with measurable outcomes and published pricing.

Businesses that are still debating whether AI agents are ready for real operations are now debating a question the market has already answered. The more practical question is how the agent economy is reshaping how businesses think about headcount and growth — and whether your team understands that shift yet.

---

If you want the playbook other teams are using with Claude and Codex right now, grab the free Working With Claude field guide. Download it here.