AWS open-sourced Kiro Crew, an agent workspace 39,000 Amazon developers adopted
Kiro Crew runs multiple coding agents across sessions with shared memory, scheduled jobs and reusable skills. Built inside Amazon as MeshClaw, it is now open source and runs locally or on your own infrastructure.
Image: InfoQ
Why it mattersIt comes with an OS-level sandbox, denied-by-default commands and a signed audit log. That is a different starting position from most agent tooling, which adds those later if at all.
Amazon released Kiro Crew as open source, a system for running several Kiro coding agents across sessions, tools and tasks. Renato Losio reported it for InfoQ on 30 August.
What it does
Asynchronous tasks that continue without supervision: incident investigation, ticket triage, migrations, pull request monitoring. It supports multi-session work through shared memory, reusable skills, scheduled jobs, concurrent agents and purpose-built Apps. Agents keep project context, delegate to subagents, and reach external tools over MCP and webhooks.
Coordination between agents runs on the Agent Client Protocol, and an Activity view shows each agent's plan, tool calls, approval steps and results as they happen.
It runs locally or on infrastructure you control.
Where it came from
It was built inside Amazon under the name MeshClaw. InfoQ reports it was adopted internally by more than 39,000 developers.
The three engineers behind it, Bolin Chen, Zejiang Guo and Zezhen Xu, describe the motivation without any product language: "Three of us wanted something simple that wasn't available internally, a way to kick off a task, walk away, come back to something worth reviewing, and to run several tasks at once instead of babysitting one prompt at a time."
They cite OpenClaw and self-learning agent tools as inspiration, and say they needed something that met their security requirements for internal development work.
Muhammad Ishaq, chief technology officer at Nextbridge, made the observation quoted in the InfoQ piece: 39,000 developers and 500 contributors in six months without being required to means people were solving real problems with it, and that kind of unplanned growth inside a company that size usually means the tool proved useful before any marketing started.
The security list
This is the part worth reading closely, because it is a list of things most agent tooling does not have at launch:
An operating system level sandbox. Commands denied by default. Suspicious-pattern blocking. Input validation. Sensitive-path blocking. Credential redaction. A signed audit log of every action.
A signed audit log is the specific one to notice. An audit trail the agent cannot edit is the difference between knowing what happened and believing what you are told.
The caveat from users
InfoQ reports that while developers are trying it on CI/CD migrations, Dependabot triage and long-running scheduled work, some have raised cost concerns, reporting that Crew consumes tokens significantly faster than Kiro CLI.
That is the expected result of running several agents in parallel, and it is worth measuring before rolling it out to a team. Concurrency multiplies the per-session cost by the number of sessions, and the whole point of the tool is to raise that number.
Skills built for other open, standards-based agent platforms work without modification, which lowers the cost of trying it.
Source
- AWS Open Sources Kiro Crew for Asynchronous Coding Agents, InfoQ, 30 August 2026
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.