AWS adds Kiro bots, agents with their memory and set tools
Each Kiro bot works with only the tools that its work needs and tells a developer before it gets more.
AWS introduced Kiro bots on Thursday. They add to Kiro Crew, its open source workspace that remembers context across sessions. A Kiro bot is a named agent that runs for a long time. It does one part of the work of a developer, such as new issues in a ticket queue. AWS gave no test results.
How it works
Each bot has its memory record that developers can edit. The bot starts parallel work sessions, watches them and joins the results. A bot can tell a developer that it needs more tools. The developer must give approval first.
The proof
AWS gave no numbers or test results. The claims come from Kathy Kam, the AWS director of engineering for agentic AI. Kam tells The New Stack that developers stop acting as the go between for sessions.
What is not known
Bots cannot give work to other bots. Kam says this stops context rot. The source does not say how correctly the memory works or how frequently a bot tells a human.
Insight: AWS is building agent controls at three levels
In October 2026, AWS released Strands Box, a sandbox with a policy language. Two engineers described hooks, which are control points in code. Kiro bots add limits for each bot. Kam says the important point is the tools that an agent can touch. This seems a move by AWS to hold the permission layer for agents. Teams that ship agents must check if these controls work with tools of other vendors.
Sources
Posted
