Amazon Web Services made two autonomous AI agents generally available on March 31: the AWS DevOps Agent for incident investigation and resolution, and the AWS Security Agent for on-demand penetration testing. Both operate without continuous human oversight across AWS, Azure, Google Cloud, and on-premises environments, according to AWS’s announcement.

AWS calls them “frontier agents” — a category it introduced at re:Invent to distinguish systems that work autonomously for hours or days from traditional AI assistants that respond to individual prompts. “These agents are changing the way we secure and operate software,” said Swami Sivasubramanian, VP of AI at AWS, in comments reported by Help Net Security. “AWS Security Agent compresses penetration testing timelines from 2-6 weeks to 1-2 days. AWS DevOps Agent gives teams 3-5x faster incident resolution.”

The launch marks a shift from selling agent-building platforms to selling the agents themselves — pre-built, priced per minute or per task, ready to deploy into production environments.

What the DevOps Agent Does

The DevOps Agent functions as an always-on site reliability engineer, according to Forbes contributor Janakiram MSV. When an alert fires, the agent begins investigating immediately — correlating telemetry, code, and deployment data across observability tools including CloudWatch, Datadog, Dynatrace, New Relic, Splunk, and Grafana. It maps application resources and their dependencies, identifies root causes, and generates mitigation plans.

Preview customers reported up to 75% lower mean time to resolution and 94% root cause accuracy, per the AWS product page. Western Governor’s University, which deployed the agent ahead of its preview launch to support over 191,000 students on its online learning platform, reduced one production incident’s resolution time from an estimated two hours to 28 minutes, according to the AWS blog post.

SiliconANGLE reported that restaurant technology platform Zenchef used the DevOps Agent during a hackathon to identify an identity access management misconfiguration in less than 30 minutes — handling the problem autonomously so engineers could stay focused on building.

The agent also does proactive prevention: it analyzes historical incident patterns and delivers targeted recommendations to strengthen system resilience. Cross-platform support for Azure and on-premises environments arrived at GA through the Model Context Protocol integration.

What the Security Agent Does

The Security Agent tackles a different problem: the gap between how many applications organizations have and how many they actually pen-test. Most enterprises limit manual penetration testing to their most critical applications because of time and cost constraints, as Forbes detailed. SiliconANGLE put a number on the gap: even the largest enterprises historically pen-test only about 10% of their most critical applications, typically once a year.

The Security Agent works by ingesting source code, architecture diagrams, and documentation to understand how an application was designed. It identifies vulnerabilities, attempts exploitation with targeted payloads and attack chains, and validates whether they pose legitimate security risks. Unlike traditional static and dynamic scanners, it understands application context and can chain vulnerabilities into higher-severity attack paths.

Bamboo Health reported that “AWS Security Agent surfaced findings that no other tool has uncovered,” according to the AWS blog. HENNGE K.K. said the agent reduced their security lifecycle’s “typical testing duration by more than 90%.”

Amy Herzog, VP and CISO at AWS, stated in the official announcement: “We’re using Security Agent ourselves at AWS.”

The agent supports AWS, Azure, Google Cloud, and on-premises systems at launch.

Pricing and the ROI Math

AWS priced both agents to force a direct comparison with human staffing costs.

The DevOps Agent costs approximately $0.50 per minute, billed per second and only when actively running, per the AWS product page. It’s free until April 10, 2026, according to SiliconANGLE.

The Security Agent charges $50 per task-hour, with an average 24-hour evaluation costing up to $1,200, per the Forbes analysis. AWS reports some customers seeing 70-90% savings on penetration testing costs compared to manual approaches.

For context: a single manual penetration test from a third-party firm typically runs $10,000 to $50,000 and takes weeks to complete. An experienced site reliability engineer in the United States commands a salary above $150,000 annually. AWS is positioning a 24-hour automated pen test at $1,200 against a multi-week manual engagement at $10,000+, and a per-minute on-call agent against a six-figure salary.

The pricing model is significant because it converts fixed headcount costs (salaries, benefits, training) into variable operational costs (pay-per-incident, pay-per-test). For organizations running lean, that math changes the build-vs-buy calculus on core operations functions.

The Competitive Landscape

AWS is not first to market with autonomous operations agents. Microsoft shipped its Azure SRE Agent to general availability on March 10, 2026, according to Microsoft’s announcement. Microsoft disclosed that it runs more than 1,300 agents internally, has mitigated over 35,000 incidents, and saves over 20,000 engineering hours monthly using its own SRE Agent.

The Azure SRE Agent GA release included features AWS doesn’t match yet: a built-in Code Interpreter that writes and executes Python to produce PDF reports, charts, Excel workbooks, and dashboards. It also has a memory and learning system that builds a knowledge base from every interaction, meaning it gets more effective over time within a specific organization’s context.

Google Cloud has no equivalent first-party autonomous operations agent, Forbes noted. It offers the Agent Development Kit for customers to build their own, Gemini Cloud Assist for operational guidance, and agentic capabilities embedded in Google SecOps for security alert triage. The positioning difference is clear: AWS and Microsoft sell pre-built agents. Google sells the platform for customers to assemble their own.

Where AWS holds a unique position is on the security side. No hyperscaler currently offers an equivalent to the Security Agent’s autonomous penetration testing capability. Microsoft’s SRE Agent focuses on incident response and operational resilience, not offensive security testing.

Constraints Enterprise Buyers Should Know

Both agents launched in only six AWS regions: US East, US West, Europe (Frankfurt and Ireland), and Asia Pacific (Sydney and Tokyo). For organizations operating under strict regulatory frameworks — particularly in financial services, healthcare, and government — limited regional availability could be a blocker.

The DevOps Agent has a significant limitation that the marketing doesn’t emphasize: it investigates and diagnoses incidents but has limited write capabilities. It cannot directly modify infrastructure or deploy fixes. It identifies root causes and recommends actions, but a human engineer still implements the remediation. Forbes flagged this explicitly: “The agent augments engineering teams rather than replacing them outright.”

The DevOps Agent also processes inference requests across US regions regardless of the customer’s selected region, which raises data residency concerns for organizations operating under GDPR, CCPA, or sector-specific data localization requirements.

For the Security Agent, autonomous penetration testing is still a nascent category. Organizations with strict compliance requirements — SOC 2, PCI-DSS, HIPAA — may still need manual pen tests conducted by certified professionals to satisfy auditors. AWS itself acknowledged in its blog that custom Model Context Protocol server connections can introduce additional prompt injection risks, and that authorized users with access to data sources the agent consumes could embed malicious instructions.

The multicloud support is new and lacks the years of battle-testing that AWS-native integrations have undergone. Organizations considering these agents for Azure or on-premises workloads should expect a rougher initial experience than those running pure AWS environments.

What This Means for Engineering Teams

The practical question for engineering leaders is no longer “will AI agents handle operations work?” It’s “when does it make financial sense to let them?”

AWS is selling these agents at price points that make them cheaper than a single on-call rotation for many mid-size companies. A $150,000/year SRE costs roughly $12,500/month. The DevOps Agent running for 500 minutes in a month — a busy incident month — costs $250. The Security Agent running quarterly pen tests at $1,200 each replaces a $40,000 annual consulting contract.

The nuance is in what these agents can’t do. They can’t deploy fixes. They can’t make architectural decisions. They can’t negotiate with stakeholders about acceptable downtime during a major outage. They accelerate diagnosis and testing, but the judgment calls — what to fix first, whether to roll back, how to communicate with customers — remain human work.

For small-to-mid-size teams without dedicated SRE or security staff, these agents represent capabilities they couldn’t previously afford. A 20-person startup running production workloads on AWS can now have autonomous incident investigation and continuous pen testing for less than the cost of a single contractor.

For large enterprises with established DevOps and security teams, the value proposition is different: these agents handle the volume work (triaging the 3 AM alerts, testing the long tail of applications that never get pen-tested) so human engineers can focus on architecture, reliability strategy, and the incidents that require judgment.

Both hyperscalers are converging on the same thesis: pre-built autonomous agents, priced per use, sold as virtual team members. The question for Google Cloud is whether the “build your own” approach can compete when AWS and Microsoft are selling finished products. The question for every engineering leader is whether they can justify not evaluating these agents when the ROI math is this straightforward.