AWS Kiro Bots: AI Agents That Remember and Coordinate

Amazon Web Services launched Kiro bots on Thursday, enhancing its Kiro Crew development workspace with persistent AI agents. These bots are capable of overseeing recurring responsibilities, coordinating other agents, and remembering developer decisions throughout the process. This new feature addresses the challenge of managing multiple AI tasks by allowing bots to handle stalled agents and reconcile results.

Each Kiro bot possesses its own memory, conversation history, and dedicated workspace, enabling developers to manage direction and review findings without reopening individual sessions. Bots can initiate multiple sessions, track progress, and consolidate outputs, even requesting human decisions when necessary while continuing other tasks.

Developers can define schedules conversationally, allowing bots to investigate failures overnight or compile weekly summaries. The bots can maintain customized dashboards, such as a GitHub monitoring bot that categorizes issues and displays backlog changes. Permissions are granular, with bots configured for specific tasks like inspecting files without editing or requesting approval before making changes.

Kiro bots can utilize different AI models, allowing teams to assign more capable models for planning and cheaper options for repetitive checks, drawing from offerings by OpenAI, Anthropic, and open-weight alternatives. The bots can run on various hardware, including laptops and EC2 instances. AWS enters a competitive landscape with products from Cursor, Anthropic, and OpenAI offering similar features like scheduling and persistent context.

AWS has priced Kiro subscriptions starting at $20 per seat per month for teams, which includes access to Crew and other tools, with pooled usage of 500 credits per seat. Additional credits cost $0.04 each. The actual cost of using continuously scheduled bots will depend on their complexity and the chosen AI models. Enterprise teams need to establish the reliability and oversight requirements for these bots in their specific workflows, starting with narrow responsibilities and gradually expanding as results are reviewed.