Kiro Review (2026): AWS's Spec-Driven AI IDE, Tested
Hand-reviewed look at Kiro, AWS's agentic spec-driven AI IDE: how specs, hooks, and agents work, credit-based pricing from $20/month, strengths, limits, and rivals.

Kiro Review (2026): AWS's Spec-Driven AI IDE, Tested
Kiro Review (2026): AWS's Spec-Driven AI IDE, Tested
Kiro is Amazon Web Services' agentic development environment, built around a simple premise: before an AI agent writes code, it should first agree with you on what it is building. Instead of jumping from a chat prompt straight to a diff, Kiro puts a structured planning layer in the middle called a spec. This hand-reviewed look covers what Kiro is, how the spec-driven workflow actually works, current pricing, its strengths and limits, and how it compares to tools like Cursor, Amazon Q Developer, and Claude Code.
Kiro reached general availability in May 2026 and is positioned as the successor to Amazon Q Developer, which AWS has placed on an end-of-support path. For AWS-aligned teams, that transition alone makes Kiro worth understanding.
What is Kiro?
Kiro is a desktop IDE built on the same open-source Code OSS foundation as Visual Studio Code, so the editor, extensions model, and keybindings feel immediately familiar. Around that editor, AWS layers an agent and a set of features designed to make specifications, not raw prompts, the unit of work. Alongside the IDE, Kiro now ships a CLI, a web preview, a mobile preview, and Crew (long-running background agents), all drawing on the same subscription.
The headline idea is spec-driven development. When you start a feature, Kiro generates three linked artifacts: a requirements.md file (user stories and acceptance criteria), a design.md file (architecture, sequence diagrams, data flow), and a tasks.md file (a discrete, trackable implementation plan). You review and approve each phase before the agent proceeds, and tasks can run individually or all at once. When you run a whole spec, Kiro builds a dependency graph and executes independent tasks concurrently in "waves," which meaningfully cuts execution time.
You authenticate with a Google account, GitHub, an AWS Builder ID, or AWS IAM Identity Center. An AWS account is not required to get started, which lowers the barrier for individual developers.
At a glance
| Aspect | Details (August 2026) |
|---|---|
| Category | Agentic, spec-driven AI IDE (plus CLI, web, mobile) |
| Vendor | Amazon Web Services (AWS) |
| Foundation | Code OSS (VS Code-based) desktop app |
| Predecessor | Amazon Q Developer (being retired) |
| Default model | Auto (mixes frontier and specialized models) |
| Selectable models | Claude Sonnet 4.5/4.6, Opus 4.x, Haiku 4.5, open-weight models |
| Free plan | Yes, 50 credits/month, perpetual |
| Paid entry | Pro at $20/month (1,000 credits) |
| GA status | Generally available since May 2026 |
Key features
Specs. The core workflow. Specs turn a loose idea into requirements, a design doc, and a task list you can approve and track. A Quick Spec mode generates all three artifacts in one pass without approval gates for well-understood work, and dedicated Bugfix Specs focus on diagnosing a root cause and preventing regressions.
Hooks. Event-driven automations that trigger the agent on actions such as saving a file or creating a component. Teams use hooks to auto-generate tests, update documentation, or run checks so that standards are enforced without anyone remembering to ask.
Steering. Project-level files that give Kiro persistent context about your architecture, conventions, and constraints. Steering keeps the agent aligned with house style across sessions rather than relearning your codebase each prompt.
Custom agents and MCP. You can define specialized subagents with scoped, pre-approved tool permissions for tasks like security review or infrastructure work, and connect external tools and data sources through the Model Context Protocol. Agent Skills, Powers, checkpoints and rewind, and cloud sessions round out the feature set.
Model choice. By default prompts run through Auto, an agent that blends frontier models (such as Claude Sonnet) with specialized models and techniques like caching and intent detection to balance quality, latency, and cost. You can pin a specific model instead, including Claude Sonnet 4.5/4.6 and Opus variants, though premium models consume credits faster.
Pricing
Kiro uses a credit model. Every prompt, spec refinement, task execution, and hook run consumes credits, metered to two decimal places, so simple edits can cost less than one credit while a complex spec task costs more. Running a task through a pinned premium model costs more credits than running it through Auto (roughly 1.3x for Sonnet versus Auto, per AWS).
| Plan | Price (USD) | Monthly credits |
|---|---|---|
| Free | $0 | 50 |
| Pro | $20/month | 1,000 |
| Pro+ | $40/month | 2,000 |
| Pro Max | $100/month | 5,000 |
| Power | $200/month | 10,000 |
Paid plans allow add-on credits or overages at $0.04 per additional credit. Unused monthly credits do not roll over. A first-time upgrade grants a $20 credit toward the first paid month, and new users receive a batch of bonus credits usable within a short window for evaluation. GovCloud pricing runs roughly 20% higher and excludes the Free tier. Pricing was verified on kiro.dev in August 2026; credit allowances and tiers have changed before and may change again.
Strengths
The spec-driven model is Kiro's real differentiator. For non-trivial features, forcing a requirements-and-design pass before code tends to produce more predictable, reviewable output and fewer of the "confidently wrong" rewrites that plague pure vibe-coding. The three artifacts also double as living documentation and a natural hand-off point between product and engineering.
Because Kiro is Code OSS-based, teams keep a familiar editor and much of the VS Code extension ecosystem. Hooks and steering give organizations concrete levers to encode standards, which matters more at team scale than raw generation speed. Parallel task execution is a genuine time-saver on larger specs. And as an AWS product, Kiro offers enterprise controls (SSO, usage analytics, centralized billing, PrivateLink/VPC endpoints, GovCloud) that appeal to regulated buyers.
Limits
The structure that makes Kiro strong is also overhead. For quick edits, a one-line fix, or exploratory prototyping, the spec ceremony can feel heavy, though Quick Spec and a lighter chat mode mitigate this. The credit system introduces cost anxiety: it can be hard to predict how many credits a large spec will burn, and pinning premium models drains allowances quickly. Non-rolling credits reward steady use and penalize bursty workflows.
Kiro is also younger than its rivals, and the pricing structure has already been revised more than once, so buyers should treat specific credit numbers as a current snapshot rather than a guarantee. The web and mobile experiences remain in preview.
Pros and cons
Pros
- Spec-driven workflow produces reviewable requirements, design, and task artifacts
- Familiar Code OSS (VS Code-based) editor and extension compatibility
- Hooks and steering encode team standards and persistent context
- Parallel task execution speeds up larger features
- Backed by AWS with strong enterprise and compliance controls
- Access to leading models, including Claude Sonnet and Opus, via Auto
Cons
- Spec ceremony is overhead for small or exploratory changes
- Credit-based pricing makes costs hard to predict on big tasks
- Unused monthly credits do not roll over
- Premium models consume credits noticeably faster than Auto
- Younger product with pricing that has already shifted
- Web and mobile clients are still in preview
Who it's for
Kiro fits teams that value planning and repeatability over raw speed: professional and enterprise engineering groups, especially those already in the AWS ecosystem or migrating off Amazon Q Developer. It suits organizations that want AI agents to follow documented standards rather than improvise. Developers who mostly want fast inline completion or a lightweight chat assistant may find Cursor or Claude Code a better everyday fit, while Amazon Q Developer users should note that AWS is actively steering them toward Kiro.
Verdict
Kiro is one of the more opinionated entries in the agentic-IDE field, and the opinion is a good one: make the agent agree on a plan before it writes code. That discipline pays off on substantial features and in team settings where consistency and reviewability matter. The trade-offs are real overhead for small tasks and a credit model that demands attention to cost. For AWS-aligned teams and anyone drawn to structured, spec-first development, Kiro is a credible, GA-stage tool that is worth a serious trial on the free tier before committing budget.
FAQ
Is Kiro free? Yes. Kiro has a perpetual Free tier with 50 credits per month, plus paid plans from $20/month.
Do I need an AWS account? No. You can sign in with Google, GitHub, an AWS Builder ID, or IAM Identity Center; an AWS account is optional for individuals.
Which AI models does Kiro use? By default it uses Auto, which blends frontier models such as Claude Sonnet with specialized models; you can also pin Claude Sonnet 4.5/4.6, Opus variants, or open-weight models.
Is Kiro replacing Amazon Q Developer? Yes. AWS has put Amazon Q Developer on an end-of-support path and directs developers to Kiro, where its newest coding models are offered.