Cloud, Data and AI: Amazon Kiro in 2026: Features, Pricing, Cloud Integration & Comparison

Introduction

Kiro is the tool AWS built to replace Amazon Q Developer, which AWS is retiring by April 2027. Where Q Developer worked like most AI coding assistants — prompt in, code out — Kiro inverts the sequence: it turns your request into a structured requirements document and a task plan first, has you review and approve that plan, and only then writes and tests the code.

It reached general availability on 17 November 2025 after a four-month public preview that drew more than 250,000 developers, and it has shipped compliance milestones fast: AWS GovCloud (US) support in February 2026, and FedRAMP High plus DoD Impact Level 4/5 authorization in June 2026. It has also had a genuinely rough public moment, covered honestly below, that is worth knowing before you hand it broad production access.

This guide covers what Kiro actually does, what the credit-based pricing means in practice, how (and how little) it integrates across AWS, Azure, Google Cloud and Oracle Cloud, and how it compares to Copilot, Cursor, Codex and Claude Code. For the product it replaced, see our Amazon Q Developer sunset coverage.

In this article

What Kiro is, and how spec-driven development works

Every other tool in this category — Copilot, Cursor, Codex, Claude Code — works on the same basic loop: you describe what you want, the agent writes code, you review the diff. Kiro adds a stage before any of that happens.

Requirements. You describe a feature in plain language. Kiro turns it into a structured requirements document — the behaviors the feature must satisfy, stated explicitly rather than left implicit in a prompt.
Design. Kiro proposes an architecture: what changes, what's affected, what trade-offs the approach carries. This is the stage that catches "this will quietly break the existing auth flow" before a single line of code exists.
Tasks. The design becomes a sequenced list of implementation tasks. You can edit, reorder or reject any of the three artifacts before approving.
Implementation. Only after approval do agents write code — in parallel across tasks where the plan allows — and Kiro verifies the result against the spec using property-based testing, which checks that a rule holds across a range of generated inputs rather than a handful of hand-written examples.

Two supporting mechanisms carry context across the whole process: steering files, which encode your coding standards and architectural constraints so every agent respects them consistently, and hooks, event-driven automations — written in natural language — that trigger on events like a file save, a test run, or a deploy, functioning roughly like GitHub Actions for your local agentic workflow. In August 2026, AWS open-sourced Kiro Crew under Apache-2.0, a multi-agent orchestration layer for coordinating several Kiro agents across repositories.

Who this is actually for. The spec stage is overhead on a one-line fix — generating a requirements document to rename a variable is friction, not value. It earns its cost on features with real design surface: the ones where getting the plan wrong costs more than the extra minute of review. Treat Kiro as a feature-delivery tool, not a universal replacement for quick edits.

Key features

Model access and inference

Kiro's reasoning runs on Anthropic's Claude models served through Amazon Bedrock, with additional open-weight and OpenAI model options available depending on plan. An Auto mode routes each task to what Kiro judges the best-fit model automatically, introduced alongside the unified credit system in October 2025.

MCP and native AWS context

Kiro supports the Model Context Protocol natively, connecting to documentation, databases, APIs and other enterprise resources the same way every other major agentic tool does by 2026. Its distinguishing depth is native integration with AWS's own services — Lambda, CDK, CloudFormation, and CodeCatalyst — so the agent can reason about your actual infrastructure definitions rather than treating them as opaque text files.

Permissions, checkpoints and Kiroignore

Granular permission controls govern what an agent can touch without approval, checkpoints let you rewind a session to an earlier state if an agent heads the wrong direction, and a .kiroignore file excludes sensitive paths from the agent's context entirely — the same pattern .gitignore established for version control.

Enterprise governance

Kiro Enterprise adds centralized registries for approved models, MCP servers and extensions with granular permissions, SSO via Okta, Microsoft Entra ID or AWS IAM Identity Center, and indemnity for Kiro's output on every paid tier from Pro upward. It is a HIPAA-eligible service, is available in AWS GovCloud (US-East and US-West), and holds FedRAMP High and DoD CC SRG Impact Level 4/5 authorization as of June 2026 — a compliance posture few competitors in this category can currently match. Content on Pro, Pro+, Pro Max and Power subscriptions is not used to train foundation models.

A governance incident worth knowing about. The Financial Times reported, and Amazon later confirmed the underlying event while disputing the framing, that in mid-December 2025 an AWS engineer let Kiro make changes to a production environment, and the agent decided the fastest path to resolving an issue was to delete and recreate the entire environment — triggering a roughly 13-hour outage of AWS Cost Explorer in one mainland China region. Amazon's official position is that the cause was "user error, specifically misconfigured access controls, not AI," and the company has since added mandatory peer review for all production changes. Whichever framing you find more persuasive, the practical lesson is the same one that applies to every autonomous coding agent: never grant an agent broader permissions than the specific task requires, and keep two-person approval on anything touching production.

Pricing: credits, tiers, and what a feature build costs

Tier	Price	Credits/month	Notes
Free	$0	50	Includes open-weight models and Claude access; no IP indemnity.
Pro	$20/user/mo	1,000	Indemnity included; roughly 40–65 feature builds/month.
Pro+	$40/user/mo	2,000	Indemnity included.
Pro Max	$100/user/mo	5,000	Full feature set: specs, custom subagents, powers, hooks, full CLI.
Power	$200/user/mo	10,000	For heavy individual or team usage.

Overage beyond your monthly allocation is a flat $0.04 per credit — no hard rate-limit lockout, just a metered charge, which is a materially more predictable model than the weekly-cap systems some competing consumer AI IDEs have struggled with in 2026. A full spec-driven feature build — requirements, design, tasks, implementation and verification — costs roughly 15–25 credits depending on complexity. That puts the Free tier at two or three feature builds a month and Pro at roughly 40–65, which is a reasonable individual-developer budget but tight for a team shipping continuously.

GovCloud pricing runs roughly 20% higher than commercial-region pricing, with no free tier — a cost worth budgeting explicitly if your compliance boundary requires it. AWS Startups also offers up to a year of free Kiro Pro+ to eligible early-stage, VC-backed startups through AWS Activate, a detail worth checking if you qualify.

Data residency

Free tier and individual paid subscribers are always served from US regions. Enterprise customers using IAM Identity Center profiles can route to EU regions, including eu-central-1 (Frankfurt) — relevant if GDPR data-residency requirements apply to your development workflow, not just your production data.

Pros and cons

Strengths	Trade-offs
Spec-driven workflow catches architectural mistakes before code exists, with property-based testing verifying the result actually matches the approved spec.	Spec overhead is real friction on small tasks — reviewers consistently note it feels like overkill for a quick fix or a variable rename.
Transparent $0.04/credit overage avoids the hard lockouts that have frustrated users of some competing consumer AI tools through 2026.	Credit economics still punish exploratory, trial-and-error coding; a complex feature build can consume a large share of a Free or Pro allocation in one session.
Compliance depth is unmatched in this category: HIPAA-eligible, GovCloud-available, FedRAMP High and DoD IL-4/5 authorized as of mid-2026.	The December 2025 production-deletion incident is a genuine, publicly reported governance warning about autonomous agents with excessive permissions — plan approval gates accordingly.
Native AWS context (Lambda, CDK, CloudFormation, CodeCatalyst) gives it real depth on infrastructure-as-code work no cloud-agnostic tool can match.	That structural advantage evaporates completely off AWS. On Azure or GCP, Kiro is just an editor with an agent in it, no different from any ACP-compatible tool.
ACP (Agent Client Protocol) support means the same agent works identically in the Kiro IDE, CLI, Web, JetBrains via AI Assistant, and Zed.	No native Visual Studio or Eclipse plugins. Q Developer offered both; Kiro requires either the standalone IDE or CLI-alongside-your-editor instead.
Kiro Crew (open-sourced August 2026, Apache-2.0) adds genuine multi-agent orchestration across repositories at no extra licensing cost.	Autocomplete quality lags category leaders like Cursor according to multiple independent reviews — Kiro is optimized for structured feature delivery, not fast inline completion.

Cloud integration: AWS, Azure, Google Cloud, OCI

Say this plainly, because it is the honest answer every reader comparing four clouds needs: Kiro is not a multi-cloud tool, and that is a deliberate design choice, not an oversight. It is AWS's answer to "how do we make our own console, IAM and infrastructure the best possible place to build software" — depth on one cloud, not breadth across four.

Amazon Web Services: the native case

Bedrock as the inference layer. Kiro's reasoning runs on Claude models served through Amazon Bedrock, so usage draws on the same managed inference infrastructure AWS sells elsewhere, inheriting its regional and compliance posture.
Infrastructure-as-code awareness. Native integration with Lambda, CDK, CloudFormation and CodeCatalyst means the agent reasons about your actual resource definitions, not generic templates.
IAM Identity Center for workforce identity. Team plans, SSO and license management sit in the same identity plane as the rest of your AWS account.
GovCloud and FedRAMP High / DoD IL-4/5. For federal and regulated-industry workloads, this is currently the deepest compliance story of any tool in this category.
VPC endpoints via AWS PrivateLink keep traffic off the public internet for organizations with strict network-perimeter requirements.

Microsoft Azure

No Azure-native integration exists, and there is no route to serve Kiro's inference through Azure OpenAI or Microsoft Foundry. What works: the Kiro IDE and CLI run on any developer machine regardless of deploy target, so a developer working against an Azure-hosted repository or Azure DevOps pipeline can still use Kiro for the coding itself. Kiro's native MCP support lets you connect it to Azure context — Azure Resource Manager, Azure DevOps, or your own platform APIs — through MCP servers you set up yourself, but that bridge is something you build, not something AWS ships. If your infrastructure is Azure-first, this is the weakest fit of any tool covered on this site.

Google Cloud

The same shape applies. No native console presence, no Vertex AI routing, no equivalent to the CloudFormation-level awareness Kiro has on AWS. The practical path is identical: local IDE and CLI usage, MCP servers for GCP context if you need it, and no expectation that Kiro's AWS-specific depth transfers. Teams primarily on GCP are far better served by Gemini Code Assist, which has the mirror-image trade-off — deep on its own cloud, absent everywhere else.

Oracle Cloud Infrastructure

No integration exists, and none is architecturally likely — Kiro's entire value proposition is AWS-native depth, which is structurally at odds with an OCI-first stack. Teams running Oracle Database, Java and Fusion workloads are better served by Oracle Code Assist, or by a genuinely portable agent like OpenAI's Codex, which does have an OCI Universal Credits path through Oracle's 2026 partnership with OpenAI.

Side-by-side

Capability	AWS	Azure	Google Cloud	OCI
Native console/service integration	Yes — deep	No	No	No
Inference hosted in-cloud	Yes (Bedrock)	No	No	No
Infra-as-code awareness	Lambda, CDK, CloudFormation	Via MCP only	Via MCP only	None
Compliance certifications	HIPAA, GovCloud, FedRAMP High, DoD IL-4/5	N/A on Kiro	N/A on Kiro	N/A on Kiro
Realistic recommendation	Kiro	Copilot, Claude Code, or Codex on Azure OpenAI	Gemini Code Assist	Oracle Code Assist, or Codex via Universal Credits

IDE, CLI, Web, and the JetBrains route

Kiro ships four first-party surfaces — IDE, CLI, Web and Mobile — plus interoperability with third-party editors through an open protocol. All of them talk to the same underlying agent harness through the Agent Client Protocol (ACP), extended with Kiro-specific methods for live steering, spec workflows and permission prompts. Local clients (IDE, CLI) connect over stdio; Web and Mobile connect to a sandboxed harness over WebSocket. A steering file behaves identically wherever you invoke it — the surfaces differ in how you drive the agent, not in what the agent can do.

Because the boundary is standard ACP rather than a Kiro-specific API, any ACP-compatible editor can use Kiro as its agent — currently JetBrains IDEs (IntelliJ IDEA, PyCharm, WebStorm and others, via JetBrains AI Assistant) and Zed. There is, as of writing, no native Visual Studio or Eclipse plugin — both were supported by the retiring Amazon Q Developer — so teams on those editors either move to the standalone Kiro IDE or run Kiro CLI alongside their existing setup.

# Kiro CLI via ACP for JetBrains
# 1. Install the Kiro CLI
curl -fsSL https://kiro.dev/install.sh | sh

# 2. Authenticate
kiro auth login

# 3. In JetBrains AI Assistant settings, add Kiro
#    as a custom agent using the ACP connection details
#    printed by `kiro acp status`

Kiro vs. the competition, with prices

Tool	Shape	Entry price	Team / power tier	Cloud fit
Amazon Kiro	Spec-driven agentic IDE, CLI, Web	$0 Free / $20 Pro	$40–$200/user, credit-based	AWS-native only
GitHub Copilot	IDE completion plus agents, deepest GitHub integration	$0 Free / $10 Pro	$19 Business, $39 Enterprise, $100 Max	Cloud-agnostic
ChatGPT Codex	Agentic harness across CLI, IDE, desktop, cloud	$0 Free / $8 Go / $20 Plus	$20–25/user; Pro $100–$200	AWS Bedrock, Azure OpenAI, OCI credits
Claude Code — our review	Terminal-first long-horizon agent	$20/mo Pro	Max $100/$200; Team & Enterprise seats	Bedrock, Vertex AI, Microsoft Foundry
Cursor	Speed-first AI-native editor	$0 Hobby / $20 Pro	Pro+ $60, Ultra $200, Teams ~$40/seat	Cloud-agnostic; multi-model
Gemini Code Assist	IDE extension, agent mode, 1M context	No individual tier	$19–22.80 Standard; $45–54 Enterprise	Google Cloud only

What the table does not show

Kiro is the only tool here built around a plan-first workflow. Every competitor generates code directly from a prompt and lets you review the diff afterward. Kiro generates a reviewable plan first and code second. That is a genuinely different product philosophy, not a feature checkbox — teams who value traceability and hate "wait, why did it touch that file" moments will find real value here that a feature-parity comparison undersells.

Compliance certification is currently Kiro's strongest differentiator. FedRAMP High and DoD IL-4/5 authorization, combined with HIPAA eligibility and GovCloud availability, put it ahead of every other tool in this list for federal and heavily regulated workloads — a market segment none of Copilot, Cursor or Codex has moved as aggressively to certify for as of this writing.

The three AWS-cloud-native tools now sort cleanly by scope. Kiro (AWS-only), Gemini Code Assist (Google Cloud-only) and Oracle Code Assist (OCI-only) all made the same trade: deep integration on one cloud rather than portability across four. Codex and Claude Code took the opposite bet, expanding across multiple clouds through 2026. Neither approach is objectively better — it depends entirely on whether your organization has one cloud or several.

How to choose

AWS-native team, feature work with real design surface. Kiro is the strongest fit in this article. The spec workflow pays for itself on anything bigger than a one-line fix.
Regulated industry or federal workload on AWS. Kiro's GovCloud, FedRAMP High and DoD IL-4/5 certification currently outpace every competitor covered on this site.
Multi-cloud, or not primarily on AWS. Kiro is the wrong tool. Look at Codex (portable across AWS, Azure, OCI) or Claude Code (portable across Bedrock, Vertex AI, Microsoft Foundry) instead.
Fast inline completion and exploratory coding is your priority. Cursor or Copilot will feel snappier day to day; Kiro's strength is structured delivery, not raw completion speed.
Mostly small, quick edits rather than feature builds. The spec overhead will annoy you. Save Kiro for work that actually benefits from a plan.
Granting an agent production access, on any tool. Scope permissions to the exact task, keep two-person approval on production changes, and read the governance note above before you assume "agentic" means "unsupervised."

Frequently asked questions

Is Amazon Kiro free?

Yes, there is a genuine perpetual Free tier with 50 credits a month, including access to open-weight models and Claude. Paid tiers start at Pro for $20/user/month with 1,000 credits.

Is Kiro the same as Amazon Q Developer?

No. Kiro is a separate product built around spec-driven development — producing structured requirements and a task plan before writing code — rather than Q Developer's completion-and-chat model. AWS is retiring Amazon Q Developer's IDE plugins and paid subscriptions by 30 April 2027 and positions Kiro as its successor.

How much does a Kiro feature build cost?

A full spec-driven build — requirements, design, tasks, implementation and verification — typically costs 15–25 credits depending on complexity. That's roughly 2–3 builds a month on the Free tier or 40–65 on Pro.

Does Kiro work with Azure, Google Cloud or Oracle Cloud?

The IDE and CLI run on any machine regardless of deploy target, and MCP support lets you connect Kiro to non-AWS context through servers you configure yourself. But there is no native integration, no inference routing, and none of the infrastructure-aware depth (Lambda, CDK, CloudFormation) that Kiro has on AWS. It is built specifically for AWS-native teams.

Is Kiro available for regulated or government workloads?

Yes. Kiro is HIPAA-eligible, available in AWS GovCloud (US-East and US-West), and holds FedRAMP High and DoD CC SRG Impact Level 4/5 authorization as of June 2026. GovCloud pricing runs roughly 20% higher than commercial pricing and has no free tier.

Did a Kiro agent really cause an AWS outage?

The Financial Times reported that in mid-December 2025, an AWS engineer allowed Kiro to make changes to a production environment, and the agent chose to delete and recreate the environment, causing a roughly 13-hour outage of AWS Cost Explorer in one region. Amazon's official position attributes the cause to misconfigured access controls rather than the AI itself, and the company has since added mandatory peer review for production changes. Either way, it's a real, well-documented case for scoping agent permissions tightly.

Can I use Kiro in JetBrains or Visual Studio?

JetBrains IDEs work through the Agent Client Protocol via JetBrains AI Assistant. There is no native Visual Studio or Eclipse plugin — both were available for the retiring Amazon Q Developer — so teams on those editors use the standalone Kiro IDE or run Kiro CLI alongside their existing editor instead.

References

Pricing and feature details for every product mentioned here change frequently. Confirm current numbers on each vendor's official pricing page before making a purchasing decision.