What Is Kiro? AWS's New AI Coding Platform Explained (2026 Guide)

Anastasiia Kuten Marketing & Partnerships Department Manager

May 18, 2026

AWS partner dedicated to startups

• 2000+ Clients
• 5+ Years of Experience
• $10M+ saved on AWS

AWS just made its biggest bet yet on AI-driven software development. It is called Kiro, and unless you have been deep in the AI coding tools space, you probably have not heard of it. That changes now.

Kiro is AWS’s new agentic IDE and CLI for AI-assisted software development. It runs on the latest Claude Opus 4.7 model with adaptive thinking. It is the official replacement for Amazon Q Developer, which AWS just announced is reaching end of support in April 2027. And it takes a fundamentally different approach to AI coding from Cursor, Copilot, or Claude Code.

This guide covers what Kiro actually is, how spec-driven development works, what it costs, how it compares to other AI coding tools, and whether your team should adopt it now or wait.

What Is Kiro?

Kiro is a family of AI coding products from AWS, available in three forms:

• Kiro IDE: a desktop application built on Code OSS (the open-source core of VS Code), with deep AI features baked in
• Kiro CLI: a command-line tool for terminal-driven workflows, CI/CD pipelines, and headless automation
• Kiro Cloud Agent: a fully autonomous version that runs in the cloud, kicked off through a web interface, for delegating work without keeping an editor open

All three share the same underlying engine, the same agent capabilities, and the same access to foundation models through Amazon Bedrock. You can start a feature in the IDE, hand it off to the cloud agent, and finish review in the CLI without losing context.

The name comes from the Japanese word for “crossroads” (きろ), which is exactly where AWS is positioning the product: at the intersection where traditional software development meets AI-powered acceleration.

Why AWS Built Kiro (And Killed Amazon Q Developer)

To understand Kiro, you have to understand what AWS is trying to fix.

The first generation of AI coding tools (Copilot, the original Q Developer, basic Cursor) focused on autocomplete and chat. Type a comment, get a function. Ask a question, get an answer. This worked for small tasks but failed on real projects.

The second generation (Claude Code, Cursor 3, Windsurf, modern Copilot) added agentic behavior. The AI could read your codebase, edit multiple files, run commands, and complete multi-step tasks. This was a huge jump, but it created a new problem: the AI now had more power but no structure. Teams ended up with what AWS calls “vibe coding chaos”: rapidly generated code with no documentation, no consistent design, no tests, and accumulating technical debt.

Kiro is AWS’s answer to vibe coding. It introduces spec-driven development as the core workflow. Instead of going straight from prompt to code, the AI first produces a structured specification: requirements, system design, and an ordered task list. You review and approve each artifact. Only then does it write code.

The bet is that developers want more structure in their AI tools, especially when shipping production software. The early signals suggest AWS may be right.

How Spec-Driven Development Actually Works

This is what makes Kiro different from every other AI coding tool. Here is the workflow in practice.

Each artifact lives in your repo. You can edit them. You can version-control them. You can hand them to teammates who do not write code. When you come back to the project in three months, the specs explain what was built and why, which is something AI-generated code rarely captures otherwise.

For quick changes that do not need the full spec workflow, Kiro added a Quick Plan mode in early 2026. This lets you do small features in a lighter workflow without abandoning the structure entirely.

Kiro’s Three Killer Features Beyond Spec-Driven Development

Spec-driven development is the headline feature, but it is not the only thing that sets Kiro apart.

1. Agent Hooks

Agent hooks are event-driven automations. They fire when a file is saved, created, deleted, or changed. The hook runs an AI agent task automatically.

Examples teams actually use:

• Save a React component → automatically update or generate the unit test
• Save an API handler → regenerate the OpenAPI spec and the TypeScript client
• Create a new Lambda function → generate the CloudFormation template for it
• Modify a database schema → update the documentation and migration script
• Save any file → run a custom security check before commit

You configure hooks once. They run silently in the background. The effect is that boring, repetitive maintenance tasks (the kind humans forget) get done automatically and consistently.

2. Steering Files

Steering files maintain persistent project context across sessions. They are markdown files in your repo that tell Kiro how to behave on this specific project. Coding standards, architectural patterns you want followed, libraries to prefer, things to avoid.

The agent reads steering files automatically. You do not have to repeat your preferences in every prompt. New team members do not have to learn them from tribal knowledge. The project’s conventions become explicit, version-controlled, and enforceable.

3. Native MCP Integration

Kiro has first-class support for the Model Context Protocol. You can connect any MCP server: the AWS MCP Server for live AWS access, GitHub MCP for repo operations, database MCPs for direct SQL queries, Cost Analysis MCP for infrastructure cost estimation, and dozens of others.

This matters because Kiro can interact with your real environment, not just generate code in isolation. It can query your database to understand the schema, check current AWS resource configurations, look up tickets in Jira, and run health checks against staging.

Kiro Pricing (As of May 2026)

Kiro moved out of free preview on October 1, 2025, and now uses a credit-based pricing model. Different models consume different amounts of credits per interaction.

Credit consumption varies by model:

• Claude Haiku and Sonnet 4.6: 1x credit multiplier (baseline)
• Claude Opus 4.6 and 4.7: 2.2x credit multiplier (more capable, more expensive)
• DeepSeek v3.2: 0.25x credit multiplier
• MiniMax M2.5: 0.25x credit multiplier
• Qwen3 Coder Next: 0.05x credit multiplier (very cheap, smaller model)

For most teams, this means: a $20 Pro plan gets you roughly 450 interactions with Opus 4.7 (1,000 ÷ 2.2), 1,000 with Sonnet, or 20,000 with Qwen3. Auto mode lets Kiro pick the best model for each task to balance quality and cost.

The Models Behind Kiro

Kiro runs exclusively on models available through Amazon Bedrock. As of May 2026, that includes:

• Claude Opus 4.7 (with adaptive thinking, available since April 17, 2026): the flagship for complex agentic work, 1M token context window, available in us-east-1 and eu-central-1
• Claude Opus 4.6 (1M context, GA): still excellent for most use cases at a lower credit cost
• Claude Sonnet 4.6 (1M context): the workhorse model, fastest balance of quality and cost
• Claude Haiku: cheapest option for simple tasks
• Open-weight models: DeepSeek v3.2, MiniMax M2.5, Qwen3 Coder Next, useful for specific tasks where the credit savings matter more than peak quality

The Auto mode (default) routes each task to the most appropriate model automatically. This is more than a marketing checkbox: it materially affects your credit consumption and output quality. Most teams should leave Auto on unless they have specific reasons to pin a model.

Note that Kiro does not support GPT models, Gemini, or any model outside Bedrock. If you want multi-provider flexibility, look at Cursor or Copilot. If you are committed to Claude (or comfortable with Bedrock as your primary inference platform), Kiro is purpose-built for that world.

Kiro vs Other AI Coding Tools

The 2026 AI coding tools landscape has gotten crowded. Here is how Kiro stacks up against the main alternatives.

The right pick depends on what matters to you.

Pick Kiro if:

• You are building on AWS and want native integration (IAM Policy Autopilot, AWS MCP Server, AgentCore)
• Your team values structure and documentation over speed of prototype
• You are migrating from Amazon Q Developer (Kiro is the official path)
• You want to use Claude as your primary model
• You want a tool that scales from desktop IDE to CLI to cloud agent without changing platforms

Pick Claude Code if:

• You live in the terminal and want minimal UI overhead
• You want maximum Claude depth without IDE complexity
• You need agent teams and parallel sub-agents (available in Teams tier)

Pick Cursor if:

• You want the most polished IDE-native experience
• You want to mix Claude, GPT, and Gemini in the same workflow
• You value community size and tutorial availability

Pick GitHub Copilot if:

• Your team is deeply on GitHub and you want tight PR integration
• You want the lowest price for getting started
• You need model flexibility (GPT, Claude, Gemini, Codex) without vendor lock-in

AWS Integrations: Where Kiro Pulls Ahead

For teams building on AWS, Kiro’s native integrations are a real differentiator. No other AI coding tool comes close to this level of AWS depth.

IAM Policy Autopilot

Launched February 2026. Kiro analyzes what your application actually does and generates least-privilege IAM policies automatically. No more starting with AdministratorAccess and meaning to tighten it later. The agent looks at your API calls, identifies the minimum permissions required, and writes the policy.

AWS MCP Server Integration

Kiro can use the AWS MCP Server (which went GA in May 2026) to access every AWS service with proper authentication. Your agent can read CloudWatch logs, inspect S3 buckets, query DynamoDB, all through your normal IAM permissions.

Built on Amazon Bedrock AgentCore

Kiro itself runs on Bedrock AgentCore (AWS’s managed agent platform). This means your interactions get session isolation, audit logging through CloudTrail, and enterprise-grade security by default.

CloudFormation and CDK Awareness

Kiro understands AWS infrastructure as code natively. It can generate CDK or CloudFormation templates that follow current AWS best practices, not what your training data knew about in 2024.

Amazon Connect, Lambda, DynamoDB Patterns

AWS has invested heavily in domain-specific knowledge for its most common services. Teams have reported building production Connect AI agents (15 backend APIs in 3 days) and full Lambda + DynamoDB applications dramatically faster than with general-purpose coding tools.

Migrating from Amazon Q Developer to Kiro

If you are currently using Amazon Q Developer, this section matters.

AWS announced in late April 2026 that Q Developer reaches end of support on April 30, 2027. New sign-ups will be blocked starting May 15, 2026. Existing customers have 12 months to transition.

The migration path is straightforward in concept but requires action.

What Stays the Same

• Amazon Q Developer in the AWS Management Console (chat, recommendations, troubleshooting)
• Q Developer in AWS documentation
• Q Developer in the AWS mobile app, Slack, and Microsoft Teams

These first-party AWS experiences are not affected by the deprecation.

What Goes Away

• Amazon Q Developer IDE plugins (the VS Code, JetBrains, and Eclipse extensions)
• Paid Q Developer subscriptions
• The dedicated Q Developer agent CLI

These all migrate to Kiro by April 30, 2027.

The Migration Steps

1. Audit your current Q Developer usage. Identify which teams use it and what for.
2. Sign up for Kiro and pick the appropriate plan. Pro for individual developers, Pro+ for active users, Power for teams running multiple agents.
3. Install the Kiro IDE alongside your current setup. Run both in parallel for a few weeks while teams learn the new workflow.
4. Migrate any custom configurations: Q Developer customization files, project-specific prompts, organizational guardrails.
5. Train teams on spec-driven development. This is the biggest culture shift, not the tool itself.
6. Decommission Q Developer once everyone is comfortable on Kiro.

Budget for the transition: most teams should plan 2 to 4 weeks of overlapping use, training time, and updating internal docs. Larger enterprises with custom Q Developer integrations need more time.

What People Are Actually Building with Kiro

The use cases that show up most often in real adoption.

Production AWS applications: Lambda functions, Step Functions workflows, DynamoDB schemas, API Gateway endpoints. Kiro’s CDK and CloudFormation awareness shortens the cycle from idea to deployed infrastructure.

Amazon Connect AI agents: AWS has published case studies showing teams building production contact center AI agents in days instead of weeks.

Internal tools and dashboards: spec-driven development is well-suited to internal apps where requirements need to be captured explicitly for handoff or maintenance.

Migrations and refactors: Kiro’s ability to read whole codebases and produce a plan before executing makes it especially useful for large refactors that previously required senior engineer attention for days.

Documentation and test generation: agent hooks make it trivial to keep tests and docs in sync with code automatically.

CI/CD pipeline tasks: Kiro CLI’s headless mode (added April 2026) lets agents run in pipelines, suggesting code review comments, generating release notes, or fixing failing tests automatically.

Limitations and Honest Concerns

No tool is perfect. Here is where Kiro is weaker than its competitors.

Model lock-in. Kiro runs exclusively on models available through Amazon Bedrock. If you want to compare a Kiro response against GPT-5 or Gemini, you have to switch tools entirely. Cursor and Copilot offer that flexibility natively.

Smaller community. Cursor and Copilot have years of head start, with more tutorials, more shared configurations, more community plugins, and more answered Stack Overflow questions. Kiro’s ecosystem is growing fast but is still smaller.

Learning curve for the spec workflow. Developers used to freeform chat-style AI assistance often find spec-driven development overly structured for quick tasks. Quick Plan mode helps but does not fully solve this. Teams need 1 to 2 weeks to get comfortable.

Credit consumption with premium models. Opus 4.7 has a 2.2x credit multiplier. A Pro plan’s 1,000 credits can disappear quickly if you stay on the best model. Plan accordingly or rely on Auto mode to route lighter tasks to cheaper models.

Region limitations for newest models. Claude Opus 4.7 inference runs in us-east-1 and eu-central-1. If you are working from other regions, expect higher latency or restricted model availability for the latest releases.

Should Your Team Adopt Kiro Now?

Decision framework:

Adopt now if:

• You are currently on Amazon Q Developer (you need to migrate anyway, sooner is better)
• You are an AWS-heavy shop and want the deepest AWS integration available
• Your team builds production software where structure and documentation matter
• You are starting greenfield projects where the spec-driven workflow can be adopted from day one

Wait or evaluate alongside other tools if:

• You are happy with your current tool and your team does not work primarily on AWS
• You need GPT-5, Gemini, or other non-Bedrock models
• Your team strongly prefers freeform AI chat over structured workflows
• You need the largest community and ecosystem (Cursor, Copilot)

For most AWS-native teams, Kiro is worth a serious evaluation. The free tier gives you 50 interactions a month to see if the spec-driven workflow fits how your team works.

Frequently Asked Questions

Is Kiro free?
The free tier gives you 50 agent interactions per month with access to core features. Paid plans start at $20/month for Pro (1,000 credits).

What models does Kiro use?
Claude Opus 4.7 (with adaptive thinking), Opus 4.6, Sonnet 4.6, Haiku, and several open-weight models (DeepSeek v3.2, MiniMax M2.5, Qwen3 Coder Next), all through Amazon Bedrock. Auto mode picks the right model per task.

Is Kiro the same as Amazon Q Developer?
No. Kiro is the replacement for Q Developer’s IDE plugins and paid subscriptions. The Q Developer experience in the AWS Console, mobile app, and Slack/Teams is staying. The IDE-based product is being deprecated and migrating to Kiro by April 2027.

Is Kiro a real AWS service?
Kiro is an AWS product but not an AWS service in the traditional sense (no service-level access through IAM, no AWS Console page). It is a separate product with its own sign-up, billing, and authentication. AWS Builder ID, Google, GitHub, and IAM Identity Center sign-in are supported.

Can Kiro work without AWS?
Yes. Kiro works with any technology stack and any cloud provider. The AWS integrations are deepest and most polished, but you can use Kiro for non-AWS work just fine.

What languages does Kiro support?
All languages supported by VS Code (since Kiro is built on Code OSS). The agent itself has been most polished for Python, JavaScript, TypeScript, and Java, with strong support for Go, Rust, C#, and others.

Does Kiro work in JetBrains IDEs?
No. Kiro is a standalone IDE based on Code OSS. JetBrains integration (which exists in Q Developer today) does not transfer to Kiro. If you are a JetBrains shop, you need to evaluate whether to migrate to the Kiro IDE or use a different AI tool.

Is Kiro the same as Cursor?
Both are VS Code-based AI coding tools, so they look similar. The differences are deep: Kiro is spec-driven by default and tightly integrated with AWS, while Cursor is multi-model and optimized for fast freeform AI assistance.

Can I use my GitHub Copilot subscription with Kiro?
No. Kiro is a separate product with its own subscription. You can use Copilot in the Kiro IDE (since it is VS Code-based and supports VS Code extensions), but the Kiro agent itself is its own thing.

Is Kiro safe for proprietary code?
Kiro runs on Amazon Bedrock, which means your code is processed within AWS infrastructure and is not used to train models. Bedrock inherits AWS’s enterprise security and compliance posture. For sensitive workloads, use IAM Identity Center authentication for the strongest controls.

Summary

Kiro is AWS’s serious attempt to define the next generation of AI coding tools, and it represents a meaningful philosophical bet: that developers want structure, not just speed.

The spec-driven workflow takes some adjustment, but it pays off in code that is easier to maintain, document, and onboard new team members onto. The deep AWS integrations are unmatched by any other tool. The model selection through Bedrock gives you access to the best Claude models available.

If you are already on Amazon Q Developer, your migration path is clear: it is Kiro, and the deadline is April 2027. If you are evaluating AI coding tools for an AWS-heavy team, Kiro deserves a serious look. If you need multi-model flexibility or work primarily outside AWS, Cursor or Copilot might suit you better.

The free tier costs nothing to try. Most teams should at least put a few hours into evaluating the spec-driven workflow before locking in their AI coding tool choice for the next year.