Skip to main content

Kiro AI (Dec 2025)

Kiro AI is AWS's AI-first IDE and agent platform (introduced mid-2025, GA announced Nov–Dec 2025). It pairs a spec-driven workflow with multi-agent automation so you can move from prompt → spec → code → tests with cloud integrations baked in.

What it is​

  • Spec-driven editor: Generates and maintains technical blueprints, keeps specs and code diffs in sync, and surfaces drift.
  • Context-aware agents: Reads repo, UI mocks, and diffs to suggest implementations and tests.
  • Mission Control Protocol (MCP): Built-in connectors to AWS services (deploy, logs, docs), plus third-party MCP servers.
  • Agent hooks / automation: Event-driven actions (run tests, update docs, scan vulns) triggered by commits or file changes.
  • Kiro Powers (Dec 2025): Load task-specific steering files + MCP stacks to turn the agent into a focused expert (e.g., “Lambda hardening”, “cost audit”).
  • Team mode + CLI: Shared contexts, permissions, and a CLI for headless automation/CI usage.

What's new (Q4 2025)​

  • GA release with team workspaces and org controls.
  • CLI parity for generating specs, running agents, and piping outputs to CI.
  • Longer-running autonomous sessions that keep task state without constant supervision.
  • Amazon internal rollout guidance favoring Kiro over third-party tools (signals rapid iteration).

Pricing / availability (public info as of Dec 2025)​

  • Free / individual: Limited agent hours and local-only projects to try the workflow.
  • Team/Org: Seat-based with usage overages for cloud agent time and MCP calls (metered like other AWS AI services).
  • Runs on macOS/Windows desktop; CLI available via installers and package managers. Sign-in with AWS, GitHub, or Google.

Quick start (desktop)​

  1. Install the desktop app from kiro.dev (macOS dmg / Windows installer).
  2. Sign in (AWS or GitHub/Google) and connect a repo (GitHub, GitLab, CodeCommit).
  3. Add cloud context: pick AWS account/region, enable MCP connectors you need (e.g., CloudWatch logs, Lambda deploy).
  4. Create a Spec from a prompt; review/lock the plan.
  5. Let the agent generate code + tests; run agent hooks (lint/test/sec) before committing.
  6. Use the timeline view to compare spec vs code diffs and resolve drift.

Quick start (CLI)​

  • Install: brew install kiro (macOS) or winget install KiroAI.Kiro (Windows) or download binary from kiro.dev/downloads.
  • Examples:
    • kiro spec new --prompt "Add payments webhook handler"
    • kiro agent run --plan kiro.plan.yaml --apply
    • kiro mcp add cloudwatch then kiro logs tail --fn my-lambda
  • Works headless in CI; pass repo path and AWS creds via env/role.

When to reach for Kiro​

  • Large feature work where you want a durable spec and drift detection.
  • Cloud-heavy stacks on AWS where MCP connectors save context-hunting time.
  • Teams that need reproducible automation (tests/docs/sec scans) tied to git events.

Cautions / gaps​

  • Best experience assumes AWS-first; multi-cloud connectors exist but are newer.
  • Agent hours are metered; budget guardrails recommended for CI usage.
  • Treat autonomy as assistive: keep humans in the loop for reviews and approvals.

Resources​