> ## Content Index
> Fetch the complete content index at: https://www.cognativ.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Kiro Bots: The Defaults, Credit Math and Hosting Choices Behind AWS's New AI Coding Agents
- URL: https://www.cognativ.com/blog/aws-kiro-bots/
- Published: 2026-11-09T14:58:48.000Z
- Updated: 2026-11-09T15:58:23.000Z
- Description: AWS's Kiro bots default to Trust mode, bill every heartbeat in credits and need a running gateway. A worked look at the settings to check before a first bot.
- Author: Kevin Anderson
- Tags: News

![](https://i.ytimg.com/vi/4-LjY6wWToo/hqdefault.jpg)

Watch: why new Kiro bots start in Trust mode, what a 30-minute heartbeat does to one Pro seat's credits, and where scheduled bots need to run.

Kiro bots, the long-running AI coding agents AWS made generally available in Kiro Crew on November 5, 2026, can act without asking first from the moment they are created. Kiro's documentation says a new bot starts in Trust mode, which skips per-action approval and leaves only the team's security policies and blocked commands in the way.

Two more details sit next to the Trust default in the documentation: every scheduled run spends credits from the same plan as interactive work, and scheduled runs depend on the Kiro Crew gateway process staying up. Together, these three settings separate a bot that takes chores off a small engineering team from one that quietly drains its plan overnight.

## Kiro Bots Are Now GA in Kiro Crew, One Job per Bot

The [Kiro team's launch post](https://kiro.dev/blog/crew-bots/?ref=cognativ.com), by product lead Kyle Seaman and Helen Hasbun, describes a bot as a named agent that owns one recurring responsibility. Each bot has its own chat, private notes that grow with feedback, a work log and a dashboard. A bot can run several worker sessions in parallel, keep working while it waits for a human decision, and follow schedules. The post also suggests creating one bot for each natural handoff in a process, from issue intake to release readiness, with each bot's input from the previous step defined by the team. Its sample roles add CI build health and pull request review between those two stages, and it says bots can run on a laptop, a Mac Mini or an Amazon EC2 instance.

Kiro announced the release on its official X account the same day, with a short demo clip.

> KKiro@kirodotdev
> 
> New in kiro crew: Kiro bots 👻 Self-improving agents for distinct roles. channel watch, issue triage, deploy guard. each builds expertise with every run, with its own memory and parallel sessions, so you delegate processes instead of tasks. Keep the why with the work. Full… pic.twitter.com/0agTtjB9BT
> 
> Nov 5, 2026[View on X](https://x.com/kirodotdev/status/2118440566350369149?ref=cognativ.com)

Kiro's official launch post for bots in Kiro Crew, listing the first roles they can take on.

## New Kiro Bots Start in Trust Mode, and Memory Isolation Is Not a File Boundary

Kiro's "How bots work" documentation lists three permission modes. Normal asks for approval on actions, apart from some built-in operations. Reads lets the bot read freely but asks before any change. Trust acts without per-action approval. New bots start in Trust, and their worker sessions inherit the bot's mode. Security policies and blocked commands still apply in every mode.

The same page makes a second point that changes how a team should set things up: one bot cannot read another bot's memory, but that separation "is not a filesystem permission boundary." A review bot and a deployment bot running on the same machine under the same account can, in principle, reach the same files. What actually limits a bot is the tool list it is granted and the security policy around it.

The risk is sharpest where developers share one build server for every repository, as many SaaS companies do. On such a machine, a triage bot created with default settings and broad file access works in a different risk class from the launch post's intake bot, which labels tickets, flags duplicates and drafts reproduction steps but cannot close a ticket. The safer order is to create the bot in Reads mode, give it only the tools its job needs, and move to Trust once its work log has been reviewed for a few weeks. Deciding a bot's tool list, permission mode and stop points before it runs is the core of [AI agent development](https://www.cognativ.com/ai/agent-development/), whether the agent is built from scratch or configured inside Kiro Crew.

![Three stacked layers for Kiro bot permission modes: Normal asks before acting, Reads allows reading only, Trust acts without per-action approval.](https://storage.ghost.io/c/cc/dc/ccdca949-0fe4-4119-8990-70a44ad75060/content/images/2026/11/aws-kiro-bots-1.png)

New Kiro bots start in the Trust layer; Reads is the safer place to begin.

## How Many Credits Does a Kiro Bot Use? Pick the Heartbeat Interval First

The launch post gives no bot pricing. Kiro's pricing page says regular Kiro subscriptions can be used with Kiro Crew, and VentureBeat noted that the launch post announces no separate per-bot fee. The documentation adds that every model call spends credits, including worker activity and scheduled runs, and that a more frequent Heartbeat raises usage. Heartbeat is off by default, needs a goal, and can run every 30 minutes, 1 hour (the default once enabled), 2 hours or 4 hours.

Kiro does not publish how many credits a run consumes, so the interval is the one cost lever a team controls before the first bill. Three intervals appear in Kiro's own materials, and each sets a different per-run budget for a single Pro seat ($20 a month, 1,000 credits) if the bot did nothing else all month.

| Interval         | Where Kiro uses it                             | Runs in 30 days | Per-run budget on one Pro seat |
| ---------------- | ---------------------------------------------- | --------------- | ------------------------------ |
| Every 30 minutes | The launch post's build-health bot checking CI | 1,440           | Under 0.7 credits              |
| Every hour       | The documented Heartbeat default               | 720             | Under 1.4 credits              |
| Every 4 hours    | The Schedules tab shown in the launch post     | 180             | Under 5.6 credits              |

The gap between the first and last row is eightfold. A build-health bot that copies the launch example's pace has less than one credit per run to work with, and any worker sessions it starts draw from the same pool. Past the allowance, add-on credits cost $0.04 each. The documentation also notes that Heartbeat skips a beat while the bot's thread is busy, so real run counts can be lower. Because Heartbeat stays off until someone gives the bot a goal and switches it on, the meter starts with that decision, not with creating the bot.

![Three equal sources around one credit pool: heartbeat runs, worker sessions and interactive use all draw on the same credits.](https://storage.ghost.io/c/cc/dc/ccdca949-0fe4-4119-8990-70a44ad75060/content/images/2026/11/aws-kiro-bots-2.png)

Heartbeats, workers and interactive chats draw on the same credit allowance.

## IDC and HFS Warn of a Coordination Tax and Permission Creep

Analysts quoted by InfoWorld turned permissions and spending into explicit warnings. IDC's Dave McCarthy described the extra spending as a kind of tax on coordination, created when a supervisor bot multiplies retries and tool calls across its workers, and advised governing bots like contractors with identity scopes, budget caps and central telemetry. HFS Research's Ashish Chaturvedi warned that developers may approve new tools out of habit, that bots shaped by different people's feedback can drift apart, and that decisions get harder to trace as a bot learns over weeks. HyperFrame Research's Stephanie Walter suggested checking whether some recurring tasks are better served by ordinary automation than by an agent.

## Scheduled Kiro Bots Need an Always-On Gateway

Kiro's documentation says scheduled activity can run while a bot's chat is closed, as long as the Kiro Crew gateway is running. The launch post's flagship example, investigating overnight build failures before the workday, therefore depends on a host that stays awake overnight; a laptop that sleeps when its owner leaves cannot be relied on to run it.

That leaves two realistic hosts for a small team: a dedicated always-on machine such as the Mac Mini the post mentions, or an EC2 instance. The cloud option brings its own decisions, including instance sizing, which repositories and secrets the instance can reach, how the gateway restarts after a failure, and where bot logs go so someone can audit them later. Those are standard [DevOps consulting](https://www.cognativ.com/consulting/cloud-infrastructure/devops/) questions, and answering them before the first schedule is enabled is cheaper than finding out after a week of silent misses.

![A cycle of four steps: the schedule fires, the always-on gateway is running, the bot runs the task, and a work log entry is written before the next run.](https://storage.ghost.io/c/cc/dc/ccdca949-0fe4-4119-8990-70a44ad75060/content/images/2026/11/aws-kiro-bots-3.png)

Scheduled runs depend on a gateway that stays up overnight.

## FAQs About Kiro Bots

Setup and permission questions for teams trying their first bot.

### What is Kiro Crew?

Kiro Crew is AWS's multi-agent development workspace, released in August 2026 under the Apache 2.0 open-source license according to InfoQ. Bots are a feature inside it.

### How do you create a Kiro bot?

Open the bots workspace in the Kiro Crew sidebar, add a bot, and choose either the default agent or a custom agent that supplies its prompt, model, tools and skills.

### Which AI models can Kiro bots use?

A bot and its workers can use different models from Kiro's catalog. The free plan is limited to open-weight models and Claude Sonnet 4.5, while paid plans add premium models.

### Can a Kiro bot merge pull requests on its own?

Only if it is given that permission. The launch post's review bot can comment and request changes but not approve or merge, and Kiro suggests loosening rules gradually.

## Set the Mode, the Interval and the Host Before the Second Bot

Kiro bots arrive with memory, workers and schedules, but for a team paying per credit and sharing one build environment, three settings decide the outcome: the permission mode a bot starts in, how often its heartbeat runs, and whether the gateway stays up when nobody is at a desk. Get those right for one bot, read its work log and credit use for a week, and only then add the next.

A first Kiro bot is a small project with three deliverables: a Reads-mode permission set, a heartbeat interval sized to one seat's credits, and an always-on host with logging. Settle all three before the bot touches a repository, and if you would like them scoped with you, [start with the details of your build environment](https://www.cognativ.com/contact/).