amazon previews 3 ai agents, including ‘kiro’ that can code on its own for days

Amazon Web Services Unveils New AI Agents

Amazon Web Services (AWS) introduced three novel AI agents, termed “frontier agents,” on Tuesday. These include an agent engineered to adapt to individual work preferences and function autonomously for extended periods.

Agent Capabilities and Functionality

These agents are designed to manage diverse tasks. They encompass code creation, security protocols like code reviews, and the automation of DevOps processes, specifically preventing issues during code deployment. Currently, preview versions of these agents are accessible.

A key assertion made by AWS centers on the “Kiro autonomous agent’s” capacity to operate independently for days.

Kiro: The Autonomous Coding Agent

Kiro is a coding agent built upon AWS’s pre-existing AI coding tool, Kiro, initially revealed in July. While the original tool facilitated prototyping, its primary intention was to generate operational code—software intended for live deployment.

To ensure code reliability, the AI must adhere to a company’s established software coding standards. Kiro achieves this through a methodology known as “spec-driven development.”

During the coding process, Kiro relies on human input to confirm or correct its assumptions, effectively establishing specifications. The autonomous agent learns by observing team workflows across various tools, including analyzing existing code.

AWS states that, subsequently, it can function independently.

Autonomous Operation and Learning

“You simply assign a complex task from the backlog, and it independently determines how to accomplish the work,” declared AWS CEO Matt Garman during his keynote at AWS re:Invent on Tuesday.

He further explained, “It actively learns your preferred work style, continuously refining its understanding of your code, products, and the standards your team adheres to.”

Amazon emphasizes that Kiro maintains “persistent context across sessions.” This means it retains information and doesn’t lose track of its objectives. Consequently, it can be assigned tasks and operate autonomously for hours or even days, according to Amazon, with minimal human oversight.

Garman illustrated this with an example: updating a critical code segment utilized by fifteen corporate software components. Instead of managing each update individually, Kiro can be tasked with resolving all fifteen within a single prompt.

Completing the Suite: Security and DevOps Agents

To fully automate coding tasks, AWS developed the AWS Security Agent. This agent autonomously identifies security vulnerabilities as code is written, conducts post-implementation testing, and proposes remediation strategies.

The DevOps Agent completes the set, automatically evaluating new code for performance issues or compatibility with other software, hardware, or cloud configurations.

Comparison with Other AI Agents

Amazon’s agents are not the first to propose extended operational durations. OpenAI announced last month that GPT‑5.1-Codex-Max, its agentic coding model, is also designed for prolonged runs, up to 24 hours.

However, it remains uncertain whether the primary obstacle to widespread agent adoption is the context window—the ability to function continuously without interruption. Concerns regarding LLM hallucination and accuracy persist, often requiring developers to act as “babysitters.”

Therefore, developers frequently prefer assigning shorter tasks and verifying results promptly before proceeding.

The Future of AI Agents

Nevertheless, expanding context windows is crucial for agents to evolve into true collaborators. Amazon’s technology represents a significant advancement in this direction.