TL;DR:
- Kiro is Amazon’s AI IDE, launched in preview in mid-2026, built around a “spec-first” workflow where you define requirements before the agent generates code.
- Unlike Cursor or Windsurf, which are primarily autocomplete and inline editing tools, Kiro treats the specification document as the source of truth — the agent works from the spec, not just the codebase.
- It’s aimed at teams building production features rather than solo developers writing scripts, and it integrates natively with AWS services.
The AI coding tool landscape in 2026 is crowded. Cursor, Windsurf, Cline, GitHub Copilot, Aider, Claude Code — each occupies a slightly different niche, but they share a common workflow: you write code (or start to), the AI helps you write more of it. The coding assistant paradigm.
Kiro takes a different approach. Instead of starting with code, you start with a spec.
What Kiro Actually Does
Kiro is a VS Code fork (like Cursor and Windsurf) with a custom AI layer built by an Amazon team. The central concept is the spec document — a structured requirements file, written in natural language, that describes what a feature or component should do.
You describe the feature you want to build. Kiro turns that into a spec. From the spec, it generates a task plan. From the task plan, it writes code. You review and approve at each stage.
The workflow looks like:
- Describe what you want to build in plain English
- Kiro generates a structured spec document (requirements, data models, API contracts)
- You review and edit the spec until it’s accurate
- Kiro generates an implementation plan with discrete tasks
- You approve the plan (or adjust it)
- Kiro executes the plan, writing code file by file
- You review diffs as they land
At no point does Kiro just start generating code and hope you can salvage it. The spec is the contract, and the agent is accountable to it.
Specs and Hooks
The two core primitives in Kiro are specs and hooks.
A spec is a .kiro/specs/<feature-name>.md file that Kiro creates and maintains. It contains structured sections: requirements, data models, API design, implementation notes. You can edit specs directly, and Kiro will re-evaluate its implementation plan when you do. The spec and the code stay in sync — if Kiro detects drift between the spec and the implementation, it flags it.
A hook is a trigger that runs automatically when something happens in the project. You can configure hooks to run when files are saved, when tests pass or fail, when PRs are opened, or on a schedule. Hooks can invoke Kiro agents, run shell commands, or call external APIs. They’re the glue between Kiro’s spec-driven workflow and your existing CI/CD pipeline.
A typical example: a hook that runs after every code change to verify the implementation still matches the spec. Or a hook that updates the spec when you manually edit the code, keeping documentation in sync with reality.
AWS Integration
Kiro ships with first-party integration for AWS services. This isn’t surprising given the provenance, but it’s well-executed: you can describe an architecture using AWS services in a spec, and Kiro will generate the infrastructure code (CDK or Terraform), the application code, and the IAM policies together, correctly wired.
For teams building on AWS, this is genuinely useful. You don’t need to describe the plumbing — Kiro knows that a Lambda function triggered by an API Gateway needs an execution role, what permissions that role needs, and how to structure the SAM template. The spec focuses on what, the agent handles how.
How It Compares to Cursor and Windsurf
The comparison to Cursor and Windsurf is the obvious one, because all three are VS Code forks with AI agents. The difference is philosophy:
Cursor is an autocomplete-first tool with an agent mode that can make multi-file edits. It’s excellent for developers who think in code and want AI to fill in the gaps. The workflow is code-centric.
Windsurf (from Codeium) introduced the “Cascade” agent, which is more autonomous — it can handle multi-step tasks with less hand-holding. Still code-centric, but with better awareness of the full codebase.
Kiro starts a level above code. If you haven’t decided exactly how to implement something, Kiro helps you think it through before any code is written. The spec-first workflow is a forcing function for upfront design — which some developers find valuable and others find frustrating.
The honest framing: Kiro will feel slower than Cursor for experienced developers who know what they want to build and just need help writing it. It will feel faster for teams where misalignment between what was spec’d and what was built is a recurring problem.
Who Kiro Is For
Kiro is best suited for:
- Teams building production features where requirements need to be explicit and traceable
- AWS-native development where infrastructure and application code need to stay in sync
- Codebases where multiple engineers contribute to the same features — the spec becomes a shared source of truth
- Developers who want to understand what they’re building before they build it rather than discover the design in the code
It’s less compelling for:
- Solo developers who work fast and iterate freely — the spec workflow adds overhead
- Exploratory or experimental work where requirements are fluid
- Non-AWS environments — the AWS integration is a differentiator, but if you’re not on AWS, you lose the main advantage over Cursor
The Spec-First Argument
There’s a philosophical case for Kiro’s approach that goes beyond the tooling. The most common failure mode in AI-assisted development isn’t that the AI writes bad code — it’s that the AI writes code for the wrong thing because the requirement was ambiguous. Cursor and Windsurf both produce high-quality code; the failure is upstream.
A spec document that you’ve reviewed and approved before any code is written forces the requirement to be explicit. It also creates a record. When you’re debugging three months later and wondering why something was implemented a particular way, the spec is there.
Whether that discipline is worth the workflow overhead is a legitimate debate. But Kiro’s bet is that for production teams, it is.
Getting Started
Kiro is available in preview as a standalone installer for macOS, Windows, and Linux. It’s free during the preview period, with pricing to be announced for general availability.
On first launch, it walks you through connecting your AWS account (optional but recommended for full feature access) and configuring your preferred AI models. Kiro uses Amazon Bedrock as the backend for its agents, with Claude claude-3-7 sonnet as the default model.
The quickstart is a “build a feature” tutorial that walks through the complete spec → plan → code workflow with a sample project. It’s the best way to form an opinion quickly — the tutorial takes about 30 minutes and you’ll know whether the workflow fits how you think by the end of it.
Kiro won’t replace Cursor for developers who are happy with the code-first workflow. But for teams that want AI that plans before it builds, it’s the most coherent implementation of that idea currently available.