AWS launches autonomous agents that work for days across code, security, and ops

Anyone who has used an AI coding assistant knows the dirty secret: while typing code is faster, you often end up playing full-time babysitter. You bounce between Jira tickets, GitHub pull requests, and Slack threads, constantly re-explaining context and babysitting tasks to ensure the model doesn't hallucinate.

At re:Invent 2025, AWS unveiled what it calls frontier agents, designed to flip this dynamic. Instead of just answering prompts, these new autonomous agents can run independently for hours or days at a time to tackle end-to-end projects across software development, security, and cloud operations.

A virtual developer, pen-tester, and SRE

The lineup starts with Kiro, a persistent developer agent that plugs into Jira, Slack, and GitHub. Kiro handles multi-repository refactoring, bug triage, and test coverage improvements. Developers simply assign backlog tasks directly in GitHub, and Kiro figures out the implementation before submitting edits as reviewable pull requests.

Alongside Kiro, the AWS Security Agent turns labor-intensive penetration testing into an on-demand process across AWS, multicloud, and hybrid stacks. SmugMug staff software engineer Andres Ruiz noted that the agent caught a subtle business logic bug exposing sensitive data that traditional static tools missed, performing work typically requiring a human pentester. When it spots an issue, it provides validated findings alongside remediation code.

Rounding out the trio is the AWS DevOps Agent, an automated on-call engineer built to reduce alerts and triage incidents. By correlating telemetry from tools like Datadog, Dynatrace, New Relic, Splunk, and Amazon CloudWatch with code commits, it pinpoints root causes. AWS claims the agent has handled thousands of internal escalations with an estimated root cause identification rate topping 86%.

Why it matters

AWS is playing to its greatest strength: enterprise infrastructure and deep operational muscle. Instead of merely battling over frontier chatbots, Amazon is integrating autonomy directly into the software development lifecycle. For developers, handing off repetitive cross-repo tasks or tedious incident triage sounds like a massive quality-of-life upgrade.

Yet as TechCrunch noted during re:Invent, AWS is still fighting to prove its AI chops beyond selling chips and cloud hosting. The core challenge will be trust. An AI agent proposing pull requests is relatively low-risk, but an operational agent digging through production infrastructure is a different beast entirely. We think engineering teams should eagerly test Kiro for backlog clearing, but maintain ironclad human-in-the-loop approvals before giving operational agents unchecked freedom in live environments.