Kiro

Kiro is an agentic development environment from Amazon Web Services that, per the vendor, sets out to turn generative code output ("vibe coding") into structured, documented "agentic engineering." It is built on Code OSS, so, per the vendor, existing VS Code settings and Open VSX–compatible extensions carry over directly; beyond the IDE, Kiro is also available as a CLI, a web interface, a mobile app, and through the separate, open-source Kiro Crew project.

Spec-driven development instead of vibe coding

The centerpiece is, per the vendor, spec mode: instead of generating code right away, Kiro first translates a requirement into executable specification files for requirements, architecture/design, and sequential tasks before implementation begins. For bug fixes, the vendor states there is a dedicated bugfix-spec variant with root-cause analysis. This approach is meant to keep decisions traceable and, per the vendor, sets Kiro apart from tools that output code without accompanying planning.

Parallel agents and property-based testing

Multiple agents can, per the vendor, work on different tasks at the same time and learn from previous sessions, which matters especially in larger codebases. For quality assurance, Kiro also uses, per the vendor, property-based, fuzzing-style tests meant to surface edge cases that classic unit tests would miss.

Steering files, hooks, and MCP support

Steering files, per the vendor, automatically enforce project standards across sessions, such as coding conventions or architectural rules. Agent hooks, per the vendor, trigger automated actions on file changes, tool use, or task completion. The Model Context Protocol additionally connects external tools and APIs, letting Kiro plug into existing toolchains.

Cloud sessions and Kiro Crew

Tasks can, per the vendor, be delegated and steered remotely through isolated cloud sandboxes, so work keeps progressing beyond the local machine. With Kiro Crew, AWS has also released a separate, Apache-2.0-licensed open-source project that, per the vendor, turns individual coding agents into persistent, self-learning engineering teams: sessions survive restarts, corrections become lasting lessons, and access runs through a desktop app, a web dashboard, a CLI, or connections like Slack and Discord.

Available models

Kiro, per the vendor, gives access to several model families, including Anthropic's Claude models (Opus, Sonnet, Haiku), an OpenAI model, and open-weight options such as DeepSeek and Qwen. An "Auto" mode automatically picks between models to, per the vendor, balance quality and cost for each task.

Security, governance, and support

For enterprises, AWS points, per the vendor, to IAM and SSO integration via AWS IAM Identity Center, central governance tooling, and IP indemnity; generated content is stated, per the vendor, to remain owned by users and not to be used to improve the underlying models. Kiro is aimed, per the vendor, at developers of every experience level as well as startups, enterprises, and educational institutions with GitHub or GitLab integration.

Free – getting started at no cost

The Free plan requires, per the vendor, no paid AWS subscription, only a free AWS account to sign in, and includes a monthly allowance of usage credits covering open model weights plus a limited quota of Claude Sonnet requests – independent reviews note this covers only a handful of full spec runs per month.

Pro, Pro+, Pro Max, and Power

Above Free, Kiro offers, per the vendor, four tiered subscription levels (Pro, Pro+, Pro Max, Power) that mainly differ in their monthly credit allowance, with higher allowances also carrying higher prices. All paid tiers, per the vendor, unlock the full set of premium models and allow extra usage through paid add-on credits, with no fixed daily or weekly limits and prepaid rather than surprise overages. Kiro lists exact prices and credit amounts on its official pricing page.

Who Kiro is for

Kiro is worth it for teams and individual developers who want a traceable planning layer before implementation and who work with several agents in parallel on larger features or codebases, rather than just completing isolated snippets. Anyone looking for more inline-style code completion directly in their familiar editor may be better served by GitHub Copilot or Cursor; Kiro's strength lies in combining a documented spec phase, parallel agent work, property-based testing, and the optional open-source Kiro Crew orchestrator for engineering teams that keep running continuously. Because Kiro is built on Code OSS, existing VS Code settings and Open VSX extensions carry over, easing the switch.