Bots in Kiro Crew: delegating processes, not sessions
Kiro has made bots generally available in Kiro Crew. A bot is a named, long-running agent with its own memory and workspace, built around one recurring responsibility: it spawns and monitors parallel sessions and reports back into a single chat. The release targets developers who are tired of shepherding many parallel agent sessions by hand.
Key points
- A bot is a long-running, named agent with its own memory, notes and dashboard, built around one responsibility
- Bots spawn, monitor and redirect parallel sessions themselves and report into a single chat
- Tools and permissions are scoped per bot; a research bot can read files but not write them
- Sensitive bots such as deployment request approval before each change and for any new tool
- Memory is isolated per bot, so lessons from one workflow don't pollute another
- Model choice is per bot: a stronger model can plan while cheaper workers run routine checks
Bots in Kiro Crew organize agentic work around responsibilities instead of sessions. Each bot has a name, its own memory, notes and workspace, is built around a goal, and can open and supervise parallel sessions, pulling drifting work back to the plan and gathering results into one chat. When a decision is needed it asks while continuing the parts that don't depend on the answer, and schedules let it investigate overnight build failures or prep weekly release notes.
Tools and permissions are scoped per bot: a research bot reads files and searches the web but cannot modify anything, while a deployment bot asks for approval before each change. Memory is isolated too, so release-process lessons don't leak into research. Model choice is per bot, letting a stronger model plan while cheaper workers run routine checks.
Real-world use
- An intake bot that labels new tickets, flags duplicates and drafts repro steps for bugs
- A build-health bot that polls CI every 30 minutes and separates new failures from known flaky tests
- A code review bot that remembers past decisions so each pull request review doesn't start cold
- A release-readiness bot that checks tests and release notes, then asks for sign-off before shipping
Why it matters
You delegate whole responsibilities instead of chasing individual agent sessions, which shifts your time to setting direction and reviewing only the results that need your judgment.