AgentComparison
Kiro · product · spec-driven coding agent and IDE

Kiro

A coding agent that turns prompts into specs and tasks, then builds with parallel agents.

Group: Coding agents

Kiro turns prompts into requirements, design and sequenced tasks, then implements them with parallel agents locally or in the cloud. Its site also describes property-based tests, a headless CLI for CI and model choice including Claude, GPT and open-weight models.

Documentation researched, not testedChecked 22 October 2026Vendor documentation, not a test

The decision in plain English

Kiro combines specs and code work with credit-based pricing. The useful control question is which surface and version you use: current IDE/CLI capability rules differ from Web sandboxing and older trust settings. Individual paid accounts do not automatically get the enterprise service-improvement exclusion.

Best fit

Developers who want specs and checks before the agent writes code.

Poor fit

Anyone who needs fixed monthly pricing; usage is credit-based and heavier models cost more credits.

Documented capabilities

Vendor statements, not verified task outcomes.

  • Specs, tasks and parallel agents Source · checked 2026-10-22
  • Credit tiers from free to $200 a month Source · checked 2026-10-22

Permissions and human control

  • Current IDE and CLI documentation uses capability rules with deny, ask and allow effects. Default policy allows workspace reads and common read-only commands; other operations prompt, with protected settings and workspace-trust rules. Web uses an isolated cloud sandbox instead of the same per-action prompt model.
  • IDE 1.x replaces the older trustedCommands model and CLI V3 replaces CLI 2.x tool settings. Older security guidance still describes prefix trust and supervised file review. Check your installed version; supervised review is not a sandbox or a universal access boundary.

Documented controls do not prove that an agent always follows them.

Limits and open questions

  • Model and credit availability vary by plan and country; add-on credits cost $0.04 each on the pricing page. No promotional credit eligibility is assumed.
  • Permissions depend on surface and version. Web does not use the local IDE/CLI interactive permission model.
  • Current and older security pages use different approval descriptions; check installed-version documentation and actual rules. Supervised mode is not an isolation boundary.
  • Data-sharing opt-out does not remove abuse-detection retention. Storage location and inference location are different questions.
  • Vendor documentation only; no code-quality, command-safety or task-success tests run.

Privacy and security

Re-verified 22 October 2026 against Kiro's data-protection docs. Free Tier and individual subscribers (paid via social login or AWS Builder ID): your content - questions, inputs, and the responses and code Kiro generates - may be used for service improvement including model training, subject to your opt-out (IDE: uncheck Data Sharing and Prompt Logging > Content Collection for Service Improvement; telemetry has a separate checkbox). Free-tier inputs may also be stored up to 60 days for abuse detection. Enterprise users: content is NOT used for service improvement; admin-enabled prompt logging/usage reporting goes to a customer-owned S3 bucket. Model-specific retention requirements for abuse detection may apply across tiers. Vendor documentation, not independently audited.

Feature-by-feature: what is documented and what is not

Official sources checked 22 October 2026. A documented feature is a vendor statement, not proof of success.

FeatureEvidence statusDetail and limit
Specs and implementationDocumentedThe vendor describes requirements, design and tasks before implementation. This is a workflow claim, not proof of correctness. Source
IDE and CLI permission rulesDocumented, version-scopedCurrent rules use deny, ask and allow per capability. Workspace reads and some read-only commands are allowed by default; other operations prompt. Source
Migration from older trust settingsDocumentedIDE 1.x replaces 0.x trustedCommands. CLI V3 replaces 2.x tool settings; verify converted rules rather than assuming old behavior. Source
Web sessionsDifferent control modelWeb uses a managed cloud sandbox, not permissions.yaml or the same per-action approval prompts. Source
Supervised reviewNot a security boundaryOlder security guidance describes file review and revert, not restricted reads or network access. Current autonomy wording differs; verify version and configuration. Source
Data use by account typeDocumented distinctionFree and individual content may support service improvement/model training unless opted out; enterprise content is excluded from service improvement. Source
Abuse retention and regionsNot zero retentionFree inputs may remain up to 60 days for abuse detection; model-specific retention can apply more widely. Inference region can differ from storage. Source
Credit pricingPublished, not a bill estimateFree 50 credits; paid tiers $20/$40/$100/$200 per user per month for 1,000/2,000/5,000/10,000 credits. Add-ons $0.04 each, taxes extra. Source
Team credit poolingPublished, eligibility-scopedTeam and Enterprise credit pooling costs $20 per seat per month, with a two-seat minimum ($40/month before tax for two seats contributing 1,000 shared monthly credits). Each seat contributes 500 credits, not the 1,000 in the legacy Enterprise Pro tier. Included credits expire monthly; purchased add-ons cost $0.04 each and expire after 12 months. Pooling uses included credits first, then the add-ons nearest expiry, and does not use overages. Pooling currently requires a Kiro profile using AWS IAM Identity Center; profiles connected directly to an external identity provider cannot use it. New Enterprise customers use pooling; existing subscribers can stay on legacy tiers or opt in. Switching subscribers to sharing is irreversible. If cancellation of a legacy subscription fails during migration, it can remain active and billed. Check the purchase summary: the AWS account determines prepaid or postpaid billing. Source

How to start with Kiro narrowly

  1. Choose IDE, CLI or Web and record the installed version before following permission instructions.
  2. Use a disposable repository without production credentials. Review the default policy and keep any added allowances narrow.
  3. For local IDE/CLI, review capability rules and workspace trust. Do not copy older prefix-trust settings blindly.
  4. For Free or individual accounts, review content-sharing and telemetry settings separately; read the abuse-retention exceptions before supplying private code.
  5. Try a small change with clear acceptance tests, inspect the diff and commands, and check credit use before increasing scope.

Official product/setup source. These are documentation-based steps, not a hands-on setup test.

Common questions about Kiro

What is Kiro for?

Developers who want specs and checks before the agent writes code. The vendor describes it as: Kiro turns prompts into requirements, design and sequenced tasks, then implements them with parallel agents locally or in the cloud. Its site also describes property-based tests, a headless CLI for CI and model choice including Claude, GPT and open-weight models.

Who should look elsewhere?

Anyone who needs fixed monthly pricing; usage is credit-based and heavier models cost more credits.

How much does Kiro cost?

Individual plans, checked 5 November 2026: Free 50 credits/month; Pro $20, Pro+ $40, Pro Max $100, Power $200 per user/month. Team and Enterprise credit pooling costs $20 per seat per month, with a two-seat minimum ($40/month before tax for two seats contributing 1,000 shared monthly credits). Each seat contributes 500 credits, not the 1,000 in the legacy Enterprise Pro tier. Included credits expire monthly; purchased add-ons cost $0.04 each and expire after 12 months. Pooling uses included credits first, then the add-ons nearest expiry, and does not use overages. Legacy Enterprise tier subscriptions can retain opt-in overages. Prices exclude tax; GovCloud pricing differs.

What can it access and what needs my approval?

Current IDE and CLI documentation uses capability rules with deny, ask and allow effects. Default policy allows workspace reads and common read-only commands; other operations prompt, with protected settings and workspace-trust rules. Web uses an isolated cloud sandbox instead of the same per-action prompt model. IDE 1.x replaces the older trustedCommands model and CLI V3 replaces CLI 2.x tool settings. Older security guidance still describes prefix trust and supervised file review. Check your installed version; supervised review is not a sandbox or a universal access boundary.

What happens to my data?

Re-verified 22 October 2026 against Kiro's data-protection docs. Free Tier and individual subscribers (paid via social login or AWS Builder ID): your content - questions, inputs, and the responses and code Kiro generates - may be used for service improvement including model training, subject to your opt-out (IDE: uncheck Data Sharing and Prompt Logging > Content Collection for Service Improvement; telemetry has a separate checkbox). Free-tier inputs may also be stored up to 60 days for abuse detection. Enterprise users: content is NOT used for service improvement; admin-enabled prompt logging/usage reporting goes to a customer-owned S3 bucket. Model-specific retention requirements for abuse detection may apply across tiers. Vendor documentation, not independently audited.

Does Kiro Enterprise still use individual subscriptions?

Pooling currently requires a Kiro profile using AWS IAM Identity Center; profiles connected directly to an external identity provider cannot use it. New Enterprise customers use pooling; existing subscribers can stay on legacy tiers or opt in. Switching subscribers to sharing is irreversible. If cancellation of a legacy subscription fails during migration, it can remain active and billed. Check the purchase summary: the AWS account determines prepaid or postpaid billing.

Source · Source

Has AgentComparison tested it?

No. This record is built from the vendor's own documentation, checked 2026-10-22. Vendor claims are attributed, not confirmed.

How to evaluate Kiro yourself

  1. Pick one low-risk task that matches its documented capabilities: specs, tasks and parallel agents.
  2. Write the expected output and what counts as failure before you start.
  3. Connect the minimum access needed, and check approval settings first.
  4. Run it twice and compare, then review every action it took.
  5. Check cost after any free allowance. Unknown price is not free.

Related

Sources and history

Change history

Documentation overview expanded with current pricing/access sources. No hands-on results claimed.

Report a correction

Costs in detail: see Kiro pricing explained for tiers, credits, add-ons and rollover rules.