Kiro is AWS's agentic coding IDE, built on the open-source base of Visual Studio Code. Its core idea is to slow the agent down: before code appears, it generates a requirements document, a design document, and a task list, each reviewed and approved by a developer. That discipline shows up clearly in reviewer feedback on complex, multi-file features, but several reviewers found it heavy-handed for small fixes, and the credit-based pricing has drawn real complaints about cost predictability.
AWS's Kiro takes a different approach to AI coding: instead of one continuous conversation, it produces a written requirements document, a design document, and a task list, each one requiring your sign-off before the agent moves to the next step or starts writing code.
Describe a feature in plain language, for example, add a review system for products.
User stories and acceptance criteria are generated using EARS notation, a requirements syntax originally developed at Rolls-Royce in 2009, not something AWS invented.
In the standard workflow, you give feedback or approve each document before Kiro moves to the next one; a separate Quick Plan mode skips these gates for well-understood features.
Technical architecture: data flow, interfaces, database schema, and API endpoints.
A dependency-ordered implementation plan, with independent tasks grouped into waves that run concurrently.
Kiro implements the approved tasks and, since general availability, can generate property-based tests that run many randomized cases against the spec; AWS describes these as evidence of correctness, not formal proof.
EDITORIAL NOTEA disputed December 2025 AWS outage that press linked to a Kiro agent action (which AWS directly disputes) and a widely circulated but untraceable 88.6% SWE-bench figure are credibility-risk signals the weighted formula doesn't otherwise capture, so a small downward adjustment applies.
AI synthesis of external reviews · not on bestaiq
Synthesized from 0 external reviews. Independent signal (Trustpilot / Reddit / verified aggregators) weighted higher than commission-carrying review sites.
Kiro is built and owned by Amazon Web Services (AWS). It runs on Amazon Bedrock and entered public preview in July 2025, reaching general availability on November 17, 2025.
There is a free tier with 50 credits a month and no credit card required. Paid plans are Pro at $20 per user/month (1,000 credits), Pro+ at $40 (2,000 credits), Pro Max at $100 (5,000 credits), and Power at $200 (10,000 credits). Extra credits cost $0.04 each on paid plans.
It depends on your tier. Kiro's own FAQ and enterprise page state that Pro, Pro+, Pro Max, Power, and Enterprise content is not used to train models. Free-tier and individual-subscriber content is stored and may be used for service improvement unless you opt out in settings.
No, they are different products, but AWS is retiring Amazon Q Developer's IDE plugins and paid subscriptions in Kiro's favor. New Q Developer signups were blocked starting May 15, 2026, and full end-of-support is scheduled for April 30, 2027.
Before writing code, Kiro produces three reviewable documents: requirements.md (user stories and acceptance criteria in EARS notation), design.md (technical architecture), and tasks.md (a dependency-ordered implementation plan). In the standard workflow, you approve each document before Kiro moves to the next.
Press reports from the Financial Times and Engadget described a December 2025, 13-hour AWS Cost Explorer outage in mainland China in which a Kiro agent deleted and recreated an environment. AWS disputes that AI caused the outage, stating it resulted from misconfigured access controls, not AI.
No official SWE-bench, HumanEval, or similar standardized score has been published by AWS or Kiro. A specific figure of 88.6 percent on SWE-bench Verified has circulated online but could not be traced to any primary source and appears to be erroneous.
Kiro's spec-driven approach is a genuine structural difference from prompt-first coding agents, and reviewers working on complex features tend to like the discipline it enforces. It suits fast, small edits less well, and buyers should budget carefully given past credit-metering problems and check which tier's content gets used for training before signing up.