Amazon Unveils “Kiro” and Two Other AI Agents That Can Code, Audit and Operate Independently for Days

At​‍​‌‍​‍‌​‍​‌‍​‍‌ its re:Invent 2025 conference, Amazon Web Services (AWS) launched three new “frontier agents” – one of which named Kiro, was described as being able to code on its own for days.

These agents are designed to grasp intricately detailed tasks while only requiring minimal human supervision. The Kiro agent is concerned with programming, it learns by reading the code of the team, gets accustomed to their workflows, and is able to deliver executable code. This is different from the existing AI coding tools in which only rough prototypes are offered and human supervision is needed heavily.

Besides the coding, AWS is also developing agents for security (vulnerability scanning) and DevOps (handling code deployment, outage prevention). The ultimate goal is very ambitious: for instance, one could go to sleep and upon awakening, find out that the AI agent has fixed bugs, completed the code audit, or resolved system issues.

AWS states that these agents are endowed with long-term memory, i.e. they retain the context even from different sessions. This, in turn, enables them to handle complex workflows that are usually executed by human teams over several days or weeks.

In order to facilitate these agents, AWS has also revealed several enhancements to its infrastructure among which is the new AI training chip, Trainium 3, that brings about both better performance and energy saving. So, it clearly indicates that AWS intends to offer a fully-fledged agentic AI solution to the enterprise market rather than just prototypes for experiments.

If the company succeeds, these frontier agents will be a powerful tool for developers to change the software industry, for security to reshape how organizations monitor threats, and for DevOps teams to free up their time with AI autonomy. For enterprises, this can translate into quicker development, lesser human errors, and enhanced work ​‍​‌‍​‍‌​‍​‌‍​‍‌productivity.