Amazon Web Services made Kiro bots generally available in Kiro Crew on 5 November 2026. Each bot is a named, long-running agent with its own memory, workspace, chat and permissions. It can start several worker sessions, watch them, redirect drift and bring results back to one conversation, including on a schedule you set in chat.
For a working engineering team, the pitch is simple: stop being the dispatcher for five parallel agent sessions. Hand one recurring responsibility to a bot, constrain what it can touch, and come back for findings rather than babysitting every run.
That sits next to earlier AWS agent moves we covered, including letting agents run Step Functions workflows without rewriting them.
What a bot actually owns
According to the Kiro launch post, each bot keeps notes you can view, edit or delete. Release decisions stay out of a research bot’s context. It can ask for tools, but nothing changes until you approve.
AWS’s starter examples are narrow and ops-shaped: an intake bot that labels issues and flags duplicates without closing them; a build-health bot that checks CI about every 30 minutes; a code review bot that can request changes but cannot approve or merge; and a release-readiness bot that checks tests and notes, then asks for sign-off.
Those are configurable workflows, not proof AWS has shipped autonomous replacements for those jobs. You still define the goal, grant access and decide what “done” means. Bots can run on a laptop, a Mac Mini or an Amazon EC2 instance. The coordinating bot and its workers can use different models, so planning can stay on a stronger model while routine checks use a cheaper one.
What it costs
Kiro pricing does not add a separate per-bot fee. Individual Free includes 50 credits a month; Pro is $20 with 1,000 credits; Pro+ is $40 with 2,000; Pro Max is $100 with 5,000; and Power is $200 with 10,000. Add-on credits are $0.04 each on paid plans. Team credit pooling for two or more seats is $20 per seat per month, with 500 included credits per seat into a shared pool.
A credit is not a finished task. Kiro billing docs say simple prompts can use under one credit, while complex work and higher-cost models use more. A bot that checks CI every 30 minutes will not cost “$20 a month” in any useful sense. Cost tracks how often it runs, which model it uses and how often it retries.
InfoWorld quotes IDC’s Dave McCarthy warning about a “coordination tax” from parallel background retries and tool calls. More activity in the background is not the same as higher productivity.
Where it breaks
The launch post does not publish comparative benchmarks, measured time saved or an error rate. You will have to measure review time, unexpected actions and credit burn on your own repos.
Permissions are the other failure mode. AWS says a review bot can comment but not merge, and a deployment bot can ask before each change. That is the right default. The risk is permission creep: bots request more tools, and busy developers approve without rechecking scope.
If you are still clarifying what counts as an agent versus a chatbot, our explainer on what an AI agent is and where it fails is the short version. For coding agents that ship with full repository access, we also looked at DeepSeek’s open-source GitHub coding agent.
Perplexity has since shipped Teammates that own ongoing Max-plan roles outside a coding IDE, with the same credit-burn warning for parallel agents.
What to do this week
Pick one recurring chore that already forces you to reconnect context across sessions: overnight CI failures, issue triage or release-note checks. Create a single Kiro bot for that job, give it the minimum tools, set a schedule and review the work log for three days. Measure developer minutes spent correcting it against minutes saved, plus credits burned. Expand only after that number looks honest.
Sources
Get the weekly email
The week's most useful AI news, a tutorial and an interview.
Anna Jakovleva
Anna Jakovleva is the senior editor for AI tools at AI Observer, leading coverage of agents and automation, design and product, and the site's tutorials. She worked in operations and product teams before joining, so she judges tools by setup time, cost and where they break. Her writing is hands-on and practical.
Spotted a mistake? Tell us on the corrections page.
