Kiro
- Free: 50 credits/mo, Pro - $20/mo, Power - $200/mo
- macOS, Windows, Linux, Web, CLI, JetBrains, Zed (ACP)
- 18+ programming languages, 50+ Powers (MCP integrations)
- Zero-retention by default, FedRAMP High, AWS GovCloud
- Official successor to Amazon Q Developer (2026)
From prompt to production - with documentation at the core
Kiro is an agent-based IDE from Amazon Web Services, launched in 2026 as the official successor to Amazon Q Developer. The main difference from Cursor, Windsurf, and GitHub Copilot is spec-driven development: before coding, the tool generates structured files for requirements, system design, and tasks, turning every change into a traceable artifact.
Key Features of Kiro
- Spec-Driven Development (SDD) - generates three files from a single prompt: requirements.md (user stories + acceptance criteria), design.md (system design and data flows), and tasks.md (task list). Design errors are caught before writing hundreds of lines of code.
- Agent Hooks - event-driven automation: saving a React component updates tests, changing an API endpoint updates the README, pre-commit triggers OWASP scanning. CI/CD logic directly in the IDE without manual execution.
- Steering Files - project configuration files with constant architecture and code style context. Supports global and project levels; relevant files auto-load based on the file type being edited.
- Autonomous Agent (preview) - a background agent that picks up tasks, implements them step-by-step with checkpoints, and opens a pull request without constant developer involvement. Revert to any previous state with one click.
- AWS-native Integration - native understanding of 200+ AWS services: Lambda, CDK, CloudFormation, DynamoDB, S3, IAM. Scaffold CDK constructs and IAM policies from a single spec file without extra prompting.
- Auto-Agent with Multimodel Routing - intelligent routing between Claude Sonnet 4.5/4.6, Haiku 4.5, Opus 4.5/4.6/4.7 via Amazon Bedrock, optimizing quality, latency, and cost through intent detection and caching.
- 50+ Powers (MCP Integrations) - built-in connectors to Figma, Terraform, Stripe, Datadog, AWS HealthOmics, and other services. Powers load as needed, not cluttering the agent context.
Pros and Cons
- The only AI-IDE with native SDD workflow
- Agent Hooks - CI/CD automation within the IDE
- Deepest integration with 200+ AWS services
- Zero-retention and FedRAMP High out of the box
- Three surfaces (IDE + CLI + Web) with unified configs
- Familiar VS Code UX, Open VSX extensions
- Spec-mode is 5 times more expensive than vibe-mode in credits
- The credit system is more complex than fixed rates
- Fewer plugins than in the VS Code Marketplace
- Limited value outside the AWS ecosystem
- Young product - fewer community resources
Kiro Pricing
- 50 credits per month
- Claude Sonnet 4.6 with limitations
- IDE, CLI, and Web
- 500 bonus credits for 14 days at start
- Unavailable in AWS GovCloud
- 1,000 credits per month
- All frontier models (Opus 4.6, Sonnet 4.6, etc.)
- Autonomous Agent (preview)
- ACP-compatible IDE (JetBrains, Zed)
- Overuse $0.04 per credit
- $20 credit on first upgrade
- 2,000 credits per month
- All Pro Features
- Overuse $0.04 per credit
- 10,000 credits per month
- All Pro+ Features
- For high-load and agent sessions
Kiro vs Competitors: An Honest Comparison
Compared to Cursor - the market leader with 1M+ paying users and 8 parallel agents - Kiro lags in autocomplete speed (sub-200ms for Cursor versus standard delay for Kiro) and ecosystem breadth. However, for AWS teams, the spec-driven workflow and native understanding of CDK/CloudFormation/IAM without prompting provide a competitive edge that Cursor lacks. Both are priced at $20/month on the basic plan.
GitHub Copilot with 20M+ users covers more IDEs and languages but offers neither SDD workflow nor automatic hooks nor AWS-specific scaffolding. Copilot starts at $10/month - cheaper, but lacks agent-level automation. Windsurf Pro at $15/month offers a smoother UX but also lacks a spec-driven approach and AWS integration depth.
Who Is Kiro For
- AWS Developers and Cloud-Native Teams - native understanding of Lambda, CDK, DynamoDB, and 200+ services without extra prompts.
- Teams with Compliance Requirements - financial sector, healthcare, public sector: GovCloud, FedRAMP High, zero-retention and code traceability.
- Senior Developers and Tech Leads - spec-driven approach provides structure instead of 'vibe coding', simplifies code review and onboarding of new team members.
- DevOps Engineers and SRE - Agent Hooks automate security scanning, IaC validation, and documentation updates with every save.
- Startups on AWS - swift transition from prototype to production with automatic requirements documentation and design solutions.
How to Start with Kiro
-
1Register at kiro.dev - create an account via email or AWS account. New users receive 500 bonus credits for 14 days for full testing.
-
2Download Kiro IDE for macOS, Windows, or Linux - or open Kiro Web in a browser. Import settings from VS Code: themes and Open VSX extensions migrate automatically.
-
3Create Your First Spec - open a project, describe a feature in the chat, and choose 'spec' mode. Kiro will generate requirements.md, design.md, and tasks.md; review and edit them before starting the agent.
-
4Configure Steering Files and Hooks - add architecture conventions in the project's steering file and create the first Agent Hook (e.g., launching a linter upon saving). This investment saves time on every subsequent request.
-
5Connect Needed Powers - add Figma MCP for design validation, Terraform, or Datadog from the catalog of 50+ integrations. Powers load only when needed and do not consume context.
Frequently Asked Questions About Kiro
What is spec-driven development and why is it needed?
Before writing code, Kiro generates three files: requirements.md with user stories, design.md with system design, and tasks.md with an implementation plan. This allows for catching design errors before writing hundreds of lines of wrong code.
How does Kiro differ from Cursor?
Cursor is faster (8 parallel agents, sub-200ms autocomplete) and has a wider plugin ecosystem with 1M+ paying users. Kiro wins in development structure (SDD), AWS-native integration, and compliance capabilities (GovCloud, FedRAMP High) - both cost $20/month on the basic plan.
Is it safe to send corporate code to Kiro?
By default, the zero-retention policy via Amazon Bedrock applies: prompts and code are not stored and not used for model training. For Pro tiers and above, via AWS IAM Identity Center, content is not used to improve the service - a critical difference from some competitors.
Can Kiro be used in JetBrains?
Yes - through Kiro CLI with support for Agent Client Protocol (ACP), the service works in JetBrains, Zed, and other ACP-compatible IDEs. All steering files and MCP configurations remain unified across all surfaces.
How does the credit system work?
Credits are deducted fractionally: from 0.01 credit for a simple request to several credits for spec-mode tasks. Overuse is billed at $0.04 per credit; unused credits do not roll over to the next month, and a team pool is not provided - each developer needs a separate subscription.
Integrations and Platforms
Kiro occupies a unique niche in the AI development tool market: it's not just a smart autocomplete but a full environment with documentation at the core of the process. If your stack is tied to AWS or requires compliance documentation, start with the free plan and 500 bonus credits right away.