Introducing Kiro Crew

Kiro Crew is an AI agent system that performs tasks developers previously had to coordinate manually across various tools and sessions. It classifies ticket queues, assigns owners, investigates failures spanning multiple repositories, and continues migrations with checkpoints and retries.

Kiro Crew started as an internal Amazon side project called MeshClaw. Built on the Kiro CLI harness, the tool was designed so developers could step away after starting a task and review the results later, rather than watching a single prompt at a time.

The project expanded as Amazon internal developers added features and edge cases needed for their own workflows. Within six months, more than 39,000 Amazon developers were using it, with approximately 500 contributors deploying 597 updates and averaging 143 commits per week.

Security is structured with multi-layered defenses, including:

• OS-level sandboxing
• Default-deny command execution
• Blocking of suspicious patterns and input validation
• Blocking access to sensitive paths
• Credential redaction
• Signed audit logs recording all actions

Kiro Crew orchestrates agents behind the Agent Client Protocol (ACP), showing the entire process in real time—from task planning and parallel sub-agent execution to tool calls, approval procedures, and result integration. The Activity view displays each agent's reasoning, tool calls, and execution results as individual agent cards.