The Kiro team at AWS announced on November 5, 2026 that bots are now generally available in Kiro Crew, its open source workspace for AI coding agents. A bot is a named agent that stays around for a long time, owns one area of a developer's work and runs its own agent sessions.
The launch targets a problem many teams now hit with agentic tools. Running several agent sessions in parallel is easy, but checking which ones stalled, which ones need a decision and what each one actually did takes real time. Kiro bots try to move that coordination work from the developer to the agent itself.
What Kiro Bots Are
Each bot gets a name, an avatar and a single continuing chat. The developer sets a goal in that chat, such as tracking open issues or keeping release checks on track, and the bot keeps working toward it over days and weeks rather than for one task.
Every bot lives in its own workspace inside Kiro Crew. It holds the bot's notes, a work log and a custom dashboard, so progress is visible without opening sessions. Bots run wherever Kiro Crew runs, which per the Kiro team includes a personal laptop, a small always-on desktop machine or a cloud server on Amazon EC2.
Key Features in the Release
- Parallel session control: a bot can start several worker sessions at once, monitor them, steer any that drift from the plan, and report the combined result in its chat.
- Non-blocking questions: when one piece of work needs a human decision, the bot asks and continues with any work that can move forward without that reply.
- Schedules: developers can set recurring jobs from the chat, for example reviewing overnight CI failures or drafting release notes each Friday.
- Separate memory: each bot keeps its own notes, so lessons learned by a deployment bot do not leak into a research bot. Users can view, change or remove anything a bot remembers.
- Scoped permissions: a bot only gets the tools its job needs. It can request more, but nothing is granted until the developer approves.
- Model choice: the bot and its worker agents can run on different models, so planning can use a stronger model while routine checks use a cheaper, faster one.
- Dashboards: the developer describes the view they want, and the bot keeps a dashboard of that work up to date.
Example Bots From the Announcement
The Kiro post outlines a starter set of bots, each with different access. An intake bot reads new tickets, labels them, spots likely duplicates and writes reproduction steps for bugs. It may comment on tickets, though closing them is off limits.
A build-health bot checks the CI pipeline every 30 minutes and separates new failures from known flaky tests. A code review bot compares each pull request with earlier team decisions and may request changes, yet final approval and merging stay with people. Finally, a release-readiness bot walks the pre-release checklist and waits for human sign-off before anything ships.
Routine scheduled activity is written to the bot's work log, while findings that need a person are posted back to the chat. When the developer corrects a bot, for instance about a failure that is expected, later runs are meant to apply that correction.
What It Means for Developers
The main shift is from handing an agent a single task to handing it an ongoing responsibility. That fits work that repeats, such as triage, CI watching and release checks, better than one-off coding jobs. Teams that already run many parallel agent sessions are the most likely to see a benefit.
The permission model deserves careful attention. Start every bot with read-only or comment-only access and widen it only after you trust its results. Keeping approve and merge rights with humans, as the example review bot does, is a sensible default for any team.
Cost is the other thing to watch. A bot that spawns several worker sessions on a schedule can use far more model calls than a single chat. Using a cheaper model for worker agents and setting sensible check intervals will help keep that under control.
The memory controls are also worth using. Because you can inspect and edit a bot's notes, review them now and then to remove outdated rules before they shape new work. If you compare options, GitHub has taken a similar path with its Agentic Workflows general availability, and JetBrains users can see how GitHub Copilot works inside JetBrains Air.
How to Get Started
Install or update Kiro Crew, then open the bots workspace from the sidebar and add a bot. You can base it on the default agent or on a custom agent you already use, which brings its prompt, model, tools and skills along without touching existing chats.
The Kiro team suggests picking one narrow responsibility that you can review in a single sitting. Name the bot, describe what it should look after, try a schedule, and adjust its rules and access as you review its output. Add more bots only once the first one is giving useful results.
Conclusion
Kiro bots turn Kiro Crew from a place to run parallel agent sessions into a way to delegate whole processes to long-running agents with their own memory, schedules and limited permissions. For teams already leaning on AI coding agents, it is a practical option to try on one well-bounded job, such as CI monitoring or issue triage, before expanding further.
Sources
- Kiro Blog: Introducing bots in Kiro Crew
- Kiro Crew product page
- InfoWorld: AWS launches Kiro bots to help devs coordinate AI coding agents
See my articles more often when you search Google.
