Amazon Web Services (AWS) has rolled out preview versions of three AI agents under the banner of "Cutting-Edge Agents". Among these innovative offerings, the "Kiro Autonomous Agent" stands out. It utilizes an AI coding tool bearing the same name (a tool with an identical name) to grasp users' work preferences and operate autonomously over extended periods, even spanning days. After undergoing training by scanning pre-existing code, it gains the capability to function independently, thanks to its "cross-session persistent context" feature. This allows it to handle tasks with minimal need for human intervention.

In addition, the "AWS Security Agent" is adept at pinpointing security vulnerabilities in coding and offering remediation advice. Meanwhile, the "DevOps Agent" takes on the responsibility of evaluating the performance and compatibility of new code.

Industry experts point out that the current hurdles to the widespread implementation of AI agents do not lie in context windows. Instead, the issues of hallucination and accuracy plaguing large language models are the main obstacles, as these require developers to constantly monitor and supervise. Nevertheless, expanding context windows is still an essential prerequisite for AI agents to evolve into reliable "colleagues". Amazon's launch of these agents marks a significant leap forward in this regard.