Kiro is AWS's bet that the missing primitive in agentic coding is not a smarter agent but a clearer spec. Every project gets three markdown files, requirements, design, tasks, that the agent fills and works from. For complex multi-step features the model produces more consistent results than vibe-first peers. The 2026 pricing revision (credit-based, with spec requests 5× the cost of vibe requests) frustrated parts of the community. Worth a serious look if you're AWS-native or rigor-first; less of a fit for fast iterative work.
- Spec-driven flow, requirements / design / tasks before any code
- Built on Code OSS (VS Code foundation), familiar interface
- Native AWS integrations, Lambda, CDK, CloudFormation, Bedrock, CodeCatalyst
- Agent hooks, event-driven automations on file save, commit, etc.
- Free tier with 50 agentic requests/month, no AWS account required
- Spec requests cost 5× vibe requests, costs add up on iterative work
- 2026 pricing change reduced free allowances; community pushback
- Smaller ecosystem of plugins/skills than Cursor or Claude Code
- Spec-first model has real overhead for quick fixes or exploration
- AWS lock-in concerns for teams allergic to single-cloud coupling
Kiro is AWS’s agentic AI IDE, launched in 2025 as the successor to
Amazon Q Developer. Built on Code OSS (the open foundation under
VS Code), Kiro adds a distinctive workflow: every project starts
with three markdown files, requirements.md, design.md, and
tasks.md, that the agent fills out and works from. It’s the
clearest implementation of “spec-driven development” in this
category.
Where it fits
Kiro fits when complex features benefit from explicit reasoning before code. Spec-driven flow uses EARS notation to convert natural language into structured requirements, then derives architectural designs and implementation tasks. For ambitious multi-step features, that scaffolding pays back as fewer mid-implementation surprises and clearer audit trails.
It’s also the natural choice for AWS-native development. Kiro integrates with Lambda, CDK, CloudFormation, Bedrock, and CodeCatalyst more deeply than any other agent in this category, and AWS GovCloud regions with private endpoints make it viable for regulated workloads. For fast vibe coding and quick fixes, Cursor or Claude Code feel lighter, Kiro’s spec overhead is real and intentional.
Pricing in practice
Free includes 50 agentic requests per month with no credit card or AWS account required. Pro at $20/mo includes 225 vibe requests and 125 spec requests. Pro+ and Power ($200/mo) cover higher-volume needs. Additional usage runs at $0.04 per vibe request and $0.20 per spec request, a 5× premium for the spec-mode flow that is Kiro’s signature feature. The 2026 pricing revision (replacing the prior 1,000-interaction tier with this credit system) drew community pushback, and is a real factor in any cost model. Kiro runs on Amazon Bedrock with Claude Sonnet 4.5 as the primary model; “Auto” mode mixes frontier models with intent detection and caching.
How it compares
Cursor, Vibe-first IDE with the fastest in-editor feedback. Pick when iteration speed matters more than spec rigor.
Claude Code, Universal-surface agent, MCP and skills-first. Pick when terminal-native workflows or MCP ecosystem matter more than the spec-driven flow.
Codex, OpenAI’s multi-surface agent with cloud-first parallelization. Pick when you’re on the GPT family and want cloud-agent orchestration.
GitHub Copilot, Lowest-friction in-IDE assistant. Pick when frictionless rollout beats deep workflow opinion.
What changed recently
The 2026 pricing revision was the dominant story, moving from a 1,000-interaction free tier to a credit-based system with a 5× premium on spec requests. Agent hooks (event-driven automations that fire on file save, commit, or other events) shipped as one of Kiro’s signature differentiators. MCP support landed natively, letting Kiro reuse the broader Claude Code skill ecosystem. AWS GovCloud availability with private endpoints expanded the regulated-industry footprint, and FedRAMP High authorization is in progress. The default model is Claude Sonnet 4.5, with “Auto” mode mixing frontier models. Code OSS foundation means Open VSX plugins and VS Code settings remain compatible, the migration cost from VS Code or Cursor is intentionally low.
Sources
- Kiro homepage, kiro.dev
- Kiro Pricing, kiro.dev
- AWS Kiro Replaces Amazon Q Developer, byteiota.com
- Amazon Kiro 2026 Review: Spec-Driven IDE vs Cursor, weavai.app
- Kiro vs Cursor 2026, morphllm.com