twimco.Let’s talk ↗︎
Menu
Field notes / Reviews

Kiro bots review: a good home for recurring dev chores, once each bot has a written job

Five agent sessions running at once sounds great until you come back from a meeting and three of them are waiting on you. AWS's answer is a named bot that owns one responsibility. It's worth trying, as long as somebody writes down what that bot may and may not touch.

Editorial engraving of a bright software studio where a manager writes on a whiteboard sheet while a developer looks up from a laptop on his knees.

Kiro bots, generally available in Kiro Crew since November 5, 2026, suit small engineering teams with recurring chores like CI triage, issue intake or release notes. Each bot keeps its own memory, tools and schedule, and runs worker sessions for you. They're included in Kiro's credit-based plans, from $20 a month. Start with one read-mostly bot, write its job down first, and keep merges and deploys behind your own approval.

What AWS shipped on November 5

Anyone who has kicked off four or five agent sessions before a meeting knows how the morning goes. You come back and one is waiting on a permission you're sure you already gave, one finished but can't explain itself, and one stopped and has been sitting idle for an hour. The agents worked in parallel. You still had to babysit every one of them.

Kiro's announcement on November 5 opens with almost exactly that scene, and it's the right problem to name. Its answer is bots, now generally available in Kiro Crew. A bot is a named, long-running agent you set up for one part of your work. It has one ongoing chat with you, its own notes, its own memory and a workspace with a dashboard you shape. When there's work to do, it can start worker sessions in parallel, watch them, redirect anything that drifts and bring the results back to that one chat. If it needs a decision, it can ask while it carries on with the parts that don't depend on your answer.

Bots can also run on a schedule. Kiro's examples are investigating overnight build failures before you start your day, and drafting release notes every Friday. They run wherever you run Kiro Crew: a laptop, a Mac Mini under a desk, or an EC2 instance.

Who it is for

Small engineering teams and solo developers who already use Kiro, and who have chores that come back every week and need a bit of judgment each time. Triaging new issues. Sorting real CI failures from the flaky tests everyone already knows about. Checking that merged changes have tests and release notes before a release. These are jobs where yesterday's context matters, which is exactly what a separate memory per bot is for.

It is a poorer fit if your process changes every sprint, or if nobody on the team has time to read what a bot writes down. Kiro's own docs are refreshingly clear that not everything needs a bot. They lay out the choices: a regular chat for interactive work, a custom agent for reusable instructions, a scheduled job for fixed steps at fixed times, a bot for an ongoing goal. If a cron job would do, use the cron job.

Decision guide

Bot, or something smaller?

ConsiderationGive it a botUse a scheduled job or a normal chat
Shape of the workAn ongoing responsibility that needs judgment each timeThe same fixed steps at a fixed time, or a one-off task
MemoryPast decisions should shape the next runEvery run can start fresh without losing anything
PermissionsA narrow, stable set of tools you can nameChanges from task to task, so you want to grant it each time
Who reviewsOne person owns the bot and reads its notes weeklyWhoever started the session reads the result
Decision framework drawn from Kiro's own guidance; assess against your team's work.
A bot is only as safe as the sentence that says what it can't do.

Where it is strong

Permissions per job, not per person. This is the part to like most. Each bot gets only the tools its job needs. Kiro's examples are well chosen: a research bot that can search the web and read files but can't change them, a deployment bot that asks for approval before each change, a code review bot that can comment and request changes but can't approve or merge. A bot can ask for more tools, and nothing changes until you approve the request.

Memory you can read and delete. Each bot keeps notes on what it has learned, and you can view, edit or delete anything it remembers. Separate memory means what the deployment bot learns about your release process doesn't leak into the research bot's view of the world. That separation is a real design choice, not a footnote.

One place to look. The dashboard and the single chat replace the habit of opening five sessions to find out where things stand. Findings and questions come to the chat; routine scheduled activity goes to a work log.

Model choice per bot. You can plan on a stronger model and let worker sessions run on a cheaper one. For a bot that checks CI every 30 minutes, that matters to the bill.

Where it falls short, or needs your discipline

Credits, not a flat fee. Kiro prices by credits. As of November 9, plans run from Free with 50 credits a month, to Pro at $20 for 1,000, up to Power at $200 for 10,000, with add-on credits at $0.04 and team pooling at $20 a seat. Credits are a unit of work, and complex prompts and stronger models use more. A bot on a 30-minute schedule spends credits whether or not anything broke. Set a usage limit before you set a schedule.

Memory can be wrong and confident. A bot that learned the wrong lesson from one odd failure will keep applying it. Reading its notes is part of owning it.

Scope creep is one approval away. Kiro's post suggests you might eventually let simple changes merge on green. Maybe. But each new tool you approve widens what the bot can do while nobody's watching. Write the rule before you click yes.

It's an AWS developer product. If your team lives in GitHub's agents or Claude Code, this is a second place for agent work to happen. That's fine, but decide which tool owns which chore so two bots don't triage the same issue.

A bot is only as safe as the sentence that says what it can't do.

What to write before you create one. Five lines are enough. The goal, in one sentence. The tools it gets. What it may never do, such as close an issue, merge, deploy or touch production data. When it must ask you instead of carrying on. And the name of the person who reads its notes every week.

Sequence of work

A first month with one bot

  1. Week 1

    Write the bot's job in five lines: goal, tools, what it may never do, when it asks, who reads its notes.

  2. Week 2

    Run a read-only build-health bot on a schedule. It reports; it doesn't change files.

  3. Week 3

    Read its notes and memory. Delete anything wrong. Correct it where it chased expected failures.

  4. Week 4

    Decide whether to add one write tool, with your approval on each change, or leave it as a reporter.

Illustrative rollout, not a client screenshot or measured result.

Plain verdict

Kiro bots are a sensible step from running sessions to owning processes, and the per-bot permissions and readable memory are better thought out than most agent products at launch. Try one if you already use Kiro and have a recurring chore that eats a morning each week. Start read-only, put a credit limit on it, and give merges and deploys to a person.

Compared with a scheduled job in your CI, a bot is more flexible and more expensive to keep honest. Compared with ad hoc agent sessions, it's calmer. If your recurring work fits in a script, keep the script. Our earlier look at GitHub Agentic Workflows makes a similar point about keeping merges with people.

If you want help deciding which chores to hand to agents, writing their limits, or wiring them into your build and release process, that is part of our web application engineering work. Talk to TWIMCO.

Related work: Spin:Market

FAQ

What are Kiro Crew bots?

Kiro bots, generally available in Kiro Crew since November 5, 2026, are named, long-running agents you set up for one part of your work. Each bot keeps its own chat, notes, memory, and tools, can start worker sessions in parallel, and can run on a schedule, such as investigating overnight build failures or drafting Friday release notes.

How much do Kiro bots cost?

They're included in Kiro's credit-based plans. As of November 9, plans run from Free with 50 credits a month, to Pro at $20 for 1,000 credits, up to Power at $200 for 10,000, with add-on credits at $0.04 and team pooling at $20 a seat. A bot on a 30-minute schedule spends credits whether or not anything broke, so set a usage limit before you set a schedule.

Who are Kiro bots for?

Small engineering teams and solo developers who already use Kiro and have recurring chores that need a bit of judgment each time, such as issue intake, CI triage, or checking that merged changes have tests and release notes. Kiro's own docs say that if a cron job would do, use the cron job.

Can a Kiro bot merge or deploy on its own?

Only if you give it those tools. Each bot gets only the tools its job needs, and a bot can ask for more, but nothing changes until you approve the request. The review's advice from the public record is to start with one read-mostly bot, write what it may never do—such as close an issue, merge, deploy, or touch production data—and keep merges and deploys behind a person.

twimco.

We build alongside the people doing the work.
Bring us the problem you’re thinking about.

Related work

Spin:Market ↗︎