AWS shipped Kiro bots this week as a generally available feature inside Kiro Crew, its open-source multi-agent development environment. The pitch: stop being a dispatcher for your own AI agents and let a supervisory bot handle the coordination instead.
How Kiro Bots Work
You create a bot in the Customize section of Kiro Crew, pick a default or custom agent as the foundation, then assign it a specific responsibility: issue triage, build monitoring, release readiness, and so on. Each bot gets its own name, avatar, instructions, tools, and permissions scoped to that role.
From there, the bot can spin up multiple worker-agent sessions in parallel, monitor their progress, redirect agents that drift off track, and consolidate results. If something needs a human decision, the bot surfaces it for approval while continuing work that does not depend on that answer.
Persistent Memory via Notes
The feature that separates this from standard agent chaining is Notes: a persistent memory layer for each bot. Notes capture developer preferences, prior decisions, project requirements, and constraints accumulated through actual work and feedback. The bot carries that context into subsequent sessions without you re-explaining your setup every time.
You can inspect, edit, or delete a bot’s Notes at any point to shape its future behavior. Developers can also schedule recurring tasks through the bot’s chat, such as investigating overnight CI failures every weekday morning or preparing release notes on a set day.
AWS is also adding Dashboards to Kiro Crew, letting developers describe how they want to visualize a bot’s ongoing activity and reports.

⚠️ The Coordination Tax
Analysts are cautious. Dave McCarthy, group vice president at IDC, flags what he calls a coordination tax: parallel background retries and tool calls can compound token consumption and compute costs quickly. Stephanie Walter, practice lead of AI stack at HyperFrame Research, puts it plainly:
“More activity in the background is not, by itself, higher productivity. Ideally, they should start with one bounded process and measure the full result: work completed correctly, developer review time, and operating cost.”
Walter also raises the question of whether every recurring dev task needs an AI agent at all. Her framing is useful: an agent earns its place when a task requires investigation, interpretation, or adaptation to unfamiliar conditions. Conventional automation is simpler, cheaper, and easier to verify when conditions are predictable.
Governance to Watch
Ashish Chaturvedi of HFS Research warns about two risks as bots accumulate weeks of instructions and corrections. First, tracing why a bot made a specific decision gets harder over time. Second, different developers may end up with bots that behave differently based on what they have each approved and corrected. McCarthy adds that permission creep is a real concern: bots requesting access to additional tools, with developers habitually approving without reassessing scope. His recommendation is to treat bots like human contractors, with strict identity scopes, budget guardrails, and centralized telemetry.
Kiro Crew is open source. If parallel agent coordination is already part of your dev workflow, the Notes feature alone is worth a look.
