AWS releases Kiro Crew, an open-source framework for autonomous coding agents

Amazon has unveiled Kiro Crew, an open-source orchestration platform designed to manage multiple AI coding agents working across different sessions and external systems. The project, previously used internally at Amazon under the name MeshClaw, enables developers to assign coding-focused work to agents that operate independently—such as investigating production incidents, sorting through pull requests, managing infrastructure migrations, and monitoring repositories—while developers focus on other priorities.

How it works#

Kiro Crew coordinates agents through the Agent Client Protocol, offering real-time visibility into what each agent is doing. The platform retains context across sessions, meaning agents can pick up where they left off, and it enables agents to spawn sub-agents to parallelize work. Integration happens via MCP (Model Context Protocol) and webhooks, allowing Kiro Crew to connect with a team's existing tooling and infrastructure.

The system can run locally or on self-hosted infrastructure and works with the Kiro CLI, supporting macOS, Linux, and Windows. It's available under the Apache 2.0 license.

Built-in security from the ground up#

The engineering team emphasized that Kiro Crew includes multiple layers of security by default: OS-level sandboxing, a deny-by-default command policy, pattern-based anomaly detection, input validation, restrictions on sensitive paths, automatic redaction of credentials, and cryptographically signed audit logs tracking every action agents take.

Real-world validation through internal use#

The release draws credibility from its internal adoption—more than 39,000 developers and 500 contributors at Amazon have used Kiro Crew over the past six months without any organizational mandate, suggesting organic demand for the capability. As the development team noted, "We were inspired by the momentum of OpenClaw and various tools with self-learning agents taking over AI work, but we needed something that met our security requirements for internal development work."

Emerging use cases and trade-offs#

Early community adopters are experimenting with continuous-integration workflows, dependency management triage, scheduled background jobs, and multi-hour tasks. However, some users have flagged token consumption as a concern, noting that Crew can burn through LLM tokens more rapidly than the Kiro CLI alone.