Amazon Web Services has unveiled a new suite of artificial intelligence agents designed to tackle a persistent challenge in AI development: persistent memory and reliable autonomous work over extended periods.
The “Frontier Agents” were introduced at the company’s re:Invent conference, promising to transform DevOps and coding by allowing AI to learn human work styles and execute complex tasks for “several days” without direct human oversight.
Central to this launch is Kiro, an advanced software agent that builds upon AWS’s earlier AI Coding tools. Kiro is engineered to address the common AI limitation of “hallucination” and short-term memory loss.
Kiro operates through “Spec-driven” development, observing team workflows, scanning existing code, and validating assumptions with human input to establish team-specific standards. This process aims for a deeper understanding of human operational patterns.
A key feature is “Persistent Context,” enabling Kiro to retain information across multiple sessions. This allows it to manage and complete extensive tasks, such as implementing 15 code updates from a single command, working autonomously for hours or even days.
AWS CEO Matt Garman stated that Kiro is designed for continuous improvement, deepening its comprehension of team products and standards over time.
Supporting Kiro are two additional agents. The AWS Security Agent operates concurrently with coding, identifying vulnerabilities, performing tests, and proposing immediate fixes. The AWS DevOps Agent focuses on testing performance and system compatibility to prevent issues during code deployment.
While AWS is not the first to explore extended AI operational durations, with OpenAI previously mentioning a GPT-5.1-Codex-Max capable of 24-hour work, the emphasis now shifts beyond mere duration to the crucial factor of reliability.
Current AI systems often suffer from “hallucination,” generating erroneous information that forces developers into a “nanny” role, meticulously checking every line of code. AWS’s initiative seeks to expand AI’s memory and automation capabilities, coupled with self-learning, to transition AI from an intern to a collaborative partner for developers.