AWS has introduced three enterprise-level AI agents at re:Invent 2025, called "Frontier Agents," covering coding, security, and DevOps processes. The preview is now available globally to developers.
The most notable among them is the "Kiro Autonomous Agent." Built on the Kiro programming platform that debuted in July, this agent continuously scans code repositories and records team standards through "spec-driven development" to independently complete complex tasks. AWS CEO Matt Garman said in his keynote speech, "Just throw the requirements from the backlog to Kiro, and it can break them down, code, test, and submit merge requests on its own, running continuously for days without human intervention."
Kiro uses cross-session persistent context technology, which the company claims "will not forget the goal due to memory overflow." During a live demonstration, Garman assigned an interface change affecting 15 systems to the agent at once, and Kiro completed adaptation, testing, and documentation updates for all modules within a few hours.
To ensure fully automated delivery, AWS also released:
- AWS Security Agent: Scans code vulnerabilities in real-time and provides repair suggestions
- AWS DevOps Agent: Automatically performs performance and compatibility tests, verifying hardware/cloud configurations
Garman emphasized that the three agents can be connected into a "hands-off" software delivery chain: Kiro codes → Security Agent reviews → DevOps Agent verifies, reducing the average feature delivery cycle from weeks to days.
Although OpenAI recently claimed that its GPT-5.1-Codex-Max supports 24-hour long tasks, developers generally remain concerned about hallucinations and accuracy, forcing them to "stay by the side." AWS acknowledges that accuracy remains a challenge but states that Kiro's persistent context marks a critical step toward making "agents become colleagues." In the coming quarters, AWS will release larger context windows and traceable audit layers to further reduce the burden of developer supervision.
