AWS Unveils Three Frontier AI Agents to Automate Coding, Security, and DevOps

Amazon Web Services Launches Frontier AI Agents

At its recent re:Invent event, Amazon Web Services announced a trio of AI agents that it calls “frontier agents.” The suite includes a coding agent named Kiro, a security‑focused agent, and a DevOps agent. Each is built to handle distinct aspects of software development and operations, and all are designed to work independently for extended periods.

Kiro Autonomous Coding Agent

Kiro builds on AWS’s existing AI coding tool first introduced in July. Unlike earlier versions that were primarily used for prototyping, Kiro is intended to produce operational code ready for live deployment. It follows a methodology called “spec‑driven development,” where the AI creates specifications by receiving human instructions, confirmations, or corrections. By scanning existing codebases and observing team workflows across development tools, Kiro learns how a team prefers to work and continuously refines its understanding of the code, products, and internal standards.

The agent maintains “persistent context across sessions,” meaning it does not lose memory of ongoing tasks when switched between prompts. AWS claims Kiro can be assigned a complex backlog item—such as updating critical code used across multiple applications—and then execute the entire task autonomously, reducing the need for step‑by‑step human oversight.

Security Agent

The Security Agent operates independently to detect security issues as code is written and after it is tested. It not only flags potential vulnerabilities but also suggests remedial actions, aiming to integrate security checks directly into the development pipeline without requiring manual code reviews for every change.

DevOps Agent

The DevOps Agent automates testing for performance, compatibility, and configuration concerns. It evaluates new code against existing software, hardware, and cloud settings, helping ensure that deployments do not introduce regressions or operational problems.

Industry Context and Outlook

While AWS’s agents are not the first to claim long‑duration autonomous operation—OpenAI has recently described its GPT‑5.1‑Codex‑Max model as capable of runs up to 24 hours—AWS emphasizes the combination of persistent context, spec‑driven development, and domain‑specific automation as a unique value proposition. The announcement also acknowledges that large language models still face challenges such as hallucinations and accuracy errors, which can require developers to act as “babysitters.” Nonetheless, AWS positions the frontier agents as a substantial step toward AI‑augmented software teams that can handle complex tasks with minimal supervision.

Matt Garman, AWS’s chief executive officer, framed the agents as co‑workers that learn and adapt, promising to deepen their understanding of a team’s code and standards over time. The preview versions of the agents are already available, signaling AWS’s intent to let customers experiment with autonomous AI assistance in real‑world development environments.