Kiro Review 2026: AWS's Spec-Driven AI IDE Tested

AWS has spent years building developer tooling at the margins — linters, deployment wizards, in-IDE code completions through Amazon Q. Kiro is its first serious attempt to build an opinionated, full-stack AI IDE from the ground up. Launched internationally on May 7, 2026, Kiro is not a reskin of Amazon Q Developer. It is a purpose-built agentic IDE that sits on the Code OSS foundation (the same open-source base as VS Code), runs Claude models via Amazon Bedrock, and introduces a spec-driven development workflow unlike anything in Cursor, Windsurf, or GitHub Copilot.

After spending several weeks with Kiro across greenfield projects and existing codebases, this review covers what actually works, where the friction is, and who should consider switching.

Key Features

Spec-Driven Development

Kiro's most distinctive feature is its three-phase approach to building: Requirements, Design, and Tasks. Before writing a single line of code for a new feature, Kiro guides you through generating a requirements document using EARS notation (Easy Approach to Requirements Syntax), an architecture design doc with component diagrams and data flow, and a sequenced task list that tracks implementation progress.

The workflow sounds heavyweight, and for small edits it is. But for features that span multiple files or services, the spec phase catches ambiguity early. In testing, asking Kiro to add a payment integration triggered a design doc that immediately surfaced a question about idempotency handling that would have caused a bug two days into implementation. That kind of front-loaded clarity is the spec workflow's genuine value.

You can skip the full spec flow and use Kiro in a more conversational mode, but the IDE is clearly optimised for the structured path.

Steering Files

Kiro introduces a .kiro/steering/ directory in your repository where you store Markdown documents that tell the IDE how to behave. A steering file might define your product's purpose, the tech stack conventions you follow, or rules about file naming and test structure. Kiro reads these files contextually and applies them throughout every interaction.

This is analogous to Cursor's .cursorrules or the emerging AGENTS.md standard, which Kiro also supports for cross-tool compatibility. The difference is that steering files are scoped — you can have global steering files that apply to every project, and per-project files that override them. Teams can commit steering files to version control and enforce consistent AI behavior across everyone who opens the repo. For organisations with strict coding standards, this is genuinely useful.

Agent Hooks

Agent Hooks are event-driven automations you configure inside Kiro. A hook triggers based on file system events — when a file is saved, created, or deleted — or on manual invocation. Practical examples from testing: a hook that runs a security scan every time a new API route file is saved; a hook that regenerates documentation when a TypeScript interface changes; a hook that adds a unit test scaffold whenever a new service file is created.

Hooks eliminate the repetitive "remind the AI to do X every time I do Y" problem. They are configured in YAML and live in the .kiro/hooks/ directory, so they version-control with the project. They can be scoped to specific file patterns, which prevents noise from triggering on every save.

Powers

Powers are Kiro's composable integration bundles. Each Power packages MCP (Model Context Protocol) tools, steering files, and hooks into a single install. The catalogue currently includes 76+ Powers covering services like Stripe, Supabase, Firebase, Terraform, Figma, and Datadog.

The practical effect is that when you install the Stripe Power, Kiro gains contextual awareness of the Stripe API, loads relevant steering rules about payment best practices, and sets up hooks for common Stripe-adjacent tasks. Context loads dynamically based on which Powers are active, which keeps the base context window lean. The Powers model is the most creative part of Kiro's architecture and points toward a future where IDE tooling is as composable as npm packages.

MCP Server Support

Beyond bundled Powers, Kiro supports remote MCP servers via Streamable HTTP with one-click installation. If you have a custom internal tool or a third-party MCP server, connecting it to Kiro is straightforward. This positions Kiro well as the MCP ecosystem continues to grow.

Autopilot and Supervised Modes

Kiro offers two agent operating modes. In Supervised mode, every file change requires your approval before it is applied — the default safe path for codebases you care about. In Autopilot mode, Kiro executes sequences of changes without asking for per-action confirmation, which substantially speeds up multi-file refactors.

You can switch between modes per task, so a large refactor might run in Autopilot while a sensitive configuration change uses Supervised. A checkpoint system snapshots state before major operations, so rolling back an Autopilot session that went sideways is a single click rather than a manual git reset.

VS Code Compatibility

Because Kiro is built on Code OSS, it imports VS Code settings, keybindings, and themes directly. Extension support comes through Open VSX. Cursor and VS Code users will find the transition largely painless at the tooling level. The gap is in the Kiro-specific extension marketplace, which is newer and thinner than what Cursor or VS Code users are accustomed to.

AWS and Enterprise Integration

Kiro is the only AI IDE with GovCloud deployment support, which matters for US federal agencies and heavily regulated industries. Enterprise plans add SAML/SCIM SSO through AWS IAM Identity Center, centralized billing, and usage analytics. For AWS-native organisations, the identity and security integration alone may justify the switch.

Model Details

Kiro runs Claude models through Amazon Bedrock. The default Auto mode uses a mix of frontier models optimised for cost and performance — in practice this routes tasks to Claude Sonnet 4 and lighter models depending on complexity. Running explicitly on Sonnet 4 costs approximately 1.3× more credits than Auto mode for equivalent tasks. There is currently no option to use GPT-5, Gemini, or other non-Anthropic models.

Pricing

Kiro uses a credit-based model. All paid tiers carry a $0.04/credit overage rate.

New accounts receive 500 bonus credits valid for 14 days, and users who upgrade from Free receive a $20 credit. Credits do not roll over between billing periods.

Team plans mirror the individual tier structure and add centralised billing, usage analytics, and SSO. Fractional credit consumption is billed to 0.01 increments, so you are charged for exactly what you use rather than rounded-up blocks.

The 50-credit Free tier is enough to meaningfully evaluate the spec workflow but will run out quickly under real usage. The Pro tier at $20/month is the practical entry point for daily use; heavy users running long Autopilot sessions will find Pro+ or Pro Max more comfortable to stay within the included allocation.

Who Is Kiro Best For?

AWS-native engineering teams get the most immediate value. The GovCloud support, IAM Identity Center SSO, and native Bedrock integration remove friction that would otherwise require custom tooling.

Product engineers working on features with meaningful scope will appreciate the spec workflow. If your typical day involves building new capabilities across several files and services, the upfront requirements-to-design-to-tasks structure pays off. It is a worse fit for developers who spend most of their time on small bug fixes or single-file patches.

Teams with strong engineering standards benefit from steering files checked into version control. Enforcing consistent AI behavior across a team without a wiki full of "how to prompt Kiro" documentation is genuinely useful for growing organisations.

Regulated industries looking at AI-assisted development for the first time may find Kiro's enterprise security story the clearest path — particularly the GovCloud option, which no other AI IDE currently offers.

Kiro is a weaker fit for developers who need model flexibility (Cursor's Sonnet/GPT-5/Gemini routing is more flexible), for solo developers doing mostly small edits (the spec overhead is real), and for extension-heavy VS Code users who rely on specific marketplace plugins not yet available in Open VSX.

Alternatives to Consider

Cursor remains the category leader for model flexibility and raw inline editing speed. Its agent mode is mature, its extension ecosystem is deep, and you can route tasks to GPT-5, Gemini, or Claude depending on the job. If you do not need the spec workflow or AWS integration, Cursor is the harder competitor to beat.

Windsurf (now owned by OpenAI) is compelling for developers who want a clean, opinionated experience with strong multi-file context. Its Cascade agent is fast and the pricing is competitive. The AWS integration story is nonexistent by comparison.

GitHub Copilot is the safe enterprise choice for organisations already in the Microsoft/Azure ecosystem. It lacks the agentic depth of Kiro or Cursor but comes with the procurement familiarity that large enterprises value.

Amazon Q Developer (Kiro's predecessor) is still available and continues to receive updates. If you are already using Q Developer for inline completions and are not ready to commit to a full IDE switch, that remains a valid path. Kiro is the long-term direction, but Q Developer is not being immediately deprecated.

Final Verdict

Kiro is the most architecturally interesting AI IDE released in 2026. The spec-driven workflow, steering files, Agent Hooks, and Powers ecosystem represent a coherent vision of what agentic development looks like when you plan the system rather than bolt AI onto an existing editor.

Its weaknesses are real. The model lock to Claude via Bedrock will frustrate developers who want to route specific tasks to other frontier models. The spec workflow has meaningful overhead for quick edits. And the extension ecosystem is still catching up to Cursor's marketplace depth.

But for AWS-native teams, regulated industries, and engineers who are tired of AI coding tools that skip the design phase and jump straight to generating code, Kiro makes a compelling case. The Powers model in particular has the potential to become an ecosystem advantage as more integrations are published.

If you are evaluating AI IDEs in mid-2026 and your work involves AWS infrastructure or complex multi-file features, Kiro deserves a serious trial. Start on the free tier with the 500-credit signup bonus — that is enough to run a real spec workflow end-to-end and judge whether the approach fits how your team actually builds.

Rating: 4.3/5

FAQ

Is Kiro free to use?

Yes. Kiro has a free tier at $0/month that includes 50 credits and access to open weight models plus Claude Sonnet 4.5. New accounts also receive 500 bonus credits valid for 14 days with no credit card required.

What AI models does Kiro use?

Kiro runs Claude models through Amazon Bedrock. The default Auto mode blends frontier models for cost and performance efficiency. Users can also select Claude Sonnet 4 directly, at approximately 1.3× the credit cost of Auto mode. GPT-5 and Gemini are not currently supported.

How does Kiro differ from Amazon Q Developer?

Kiro is a purpose-built agentic IDE built on Code OSS. Amazon Q Developer was focused on inline code completions and chat. Kiro introduces the spec-driven workflow, steering files, Agent Hooks, and the Powers ecosystem — it is a substantially more ambitious product aimed at replacing your full IDE rather than augmenting an existing one.

Can I use VS Code extensions in Kiro?

Kiro supports Open VSX extensions and can import VS Code settings, themes, and keybindings. Extensions available exclusively through the VS Code Marketplace (Microsoft's proprietary marketplace) are not directly installable, though many popular extensions are available through Open VSX.

Do Kiro credits roll over?

No. Credits do not roll over between billing periods. If you consistently hit your included allocation, consider upgrading to the next tier — the $0.04/credit overage rate adds up quickly under heavy agent use.

Is Kiro suitable for enterprise teams?

Yes, particularly for AWS-native organisations. Kiro offers SAML/SCIM SSO via AWS IAM Identity Center, GovCloud deployment (unique among AI IDEs), centralised billing, and usage analytics on team plans. Steering files committed to version control also help enforce consistent AI behavior across large engineering teams.

What are Kiro Powers?

Powers are composable integration bundles that combine MCP tools, steering files, and Agent Hooks into a single install. Installing a Power (for example, the Stripe or Supabase Power) gives Kiro contextual awareness of that service's APIs and best practices, and sets up relevant automations. There are currently 76+ Powers available.