Kiro Review 2026: AWS's Spec-Driven AI IDE for Australia

Name: Kiro Review (2026): AWS's Spec-Driven AI IDE, Tested for Australia
Item: Kiro
Rating: 16
Author: Adam

16/25

★★★★☆

Very Good

Kiro · Amazon Web Services

Verified 9 Oct 2026 Reviewed 25 Jun 2026

Value for Money3/5

Ease of Use3/5

Feature Depth4/5

Australian Fit2/5

Reliability & Support4/5

From: A$28/mo · ex-GST · billed in USD (US$20)

The gate: Five stars need 21+ and Australian Fit of at least 4/5.

Skip to the verdict

Pricing figures updated 6 September 2026: USD prices are converted at the current rate (1 USD ≈ 1.40 AUD, verified 5 September 2026) and shown ex-GST, replacing the retired 1.55 + GST basis. A GST-registered Australian business that supplies its ABN is not charged GST on an imported digital service; other buyers should allow ~10% on top (Methodology).

The verdict: Kiro is Amazon's answer to "vibe coding" — an AI IDE that makes you agree a written spec before it builds anything. That discipline is the whole point: requirements, a design, then tasks, so the AI builds what you actually meant rather than guessing. It runs top-tier Claude models (Sonnet 4.6, Opus 4.8) and an Auto agent that picks models for you, it went generally available in November 2025, and it has the full weight of AWS behind it. But this is a developer tool, not a no-code builder — if you can't read code, it is not for you. Billing is in US dollars on an expiring monthly credit model, your prompts on the free and individual tiers can be used to train models unless you pay for enterprise, and your code is processed and stored in the United States with no Australian region. For Australian developers it is a genuinely strong, structured IDE. For a non-technical owner who wants an app built, look elsewhere. Score: 16/25 — Very Good.

What it is

Kiro is an agentic AI IDE built by Amazon Web Services. You can think of it as a code editor with an AI agent baked in — but its defining idea is spec-driven development. Instead of typing a vague prompt and hoping the AI guesses right ("vibe coding"), Kiro pushes you to write a structured specification first: the requirements in a formal EARS format ("THE system SHALL allow authenticated users to view active listings"), then a technical design, then a task list. The AI works from that spec, so what it builds is anchored to something you agreed up front. AWS markets this as moving "beyond AI coding to agentic engineering" — and it really was first to bring spec-first workflows to the AI-IDE pack.

The backstory is short and corporate. AWS launched Kiro in preview in July 2025 with a waitlist, and more than 100,000 developers joined to try it. It reached general availability on 17 November 2025 — the waitlist gone, plus property-based testing, checkpointing, a terminal CLI and team plans. Under the bonnet, Kiro is built on Code OSS, the same open-source core as Microsoft's VS Code (the same base Cursor and Windsurf use), so it feels instantly familiar and runs Open VSX extensions. The AI layer on top is the difference.

For an Australian buyer, the headline is the trade-off. The AWS name brings stability, indemnity and a serious security posture — but Kiro is a US-hosted developer product with US-dollar billing and no Australian footprint of its own, and we will come back to all three.

What it costs in Australia

Kiro bills in US dollars on a credit model. There are five individual tiers, and the same prices apply to team plans (per user):

Plan	USD/mo	Approx AUD/mo (ex-GST)	Credits/mo	Notes
Free	$0	$0	50	Open-weight models + Claude Sonnet 4.5, rate-limited
Pro	US$20	~A$28	1,000	Premium models, $0.04/credit overage
Pro+	US$40	~A$56	2,000	Premium models, $0.04/credit overage
Pro Max	US$100	~A$140	5,000	Premium models, $0.04/credit overage
Power	US$200	~A$280	10,000	Premium models, $0.04/credit overage
Enterprise	Custom	—	Custom	SSO, central billing, usage analytics

The credit model is the part to understand. A credit is a unit of work, charged fractionally to two decimal places — a small edit can cost less than one credit, while running a full spec task costs more. Critically, every kind of work draws from the same pool: vibe prompts, spec refinement, task execution and agent-hook runs all spend credits. The model you choose also changes the rate — running a task through Claude Sonnet 4.6 costs about 1.3× what the cheaper Auto agent costs for the same job. Credits reset monthly and do not roll over — anything unused expires. On paid plans you can switch on overage at US$0.04 per extra credit, billed at month-end, but it is off by default.

A few Australian-specific notes. Pricing is USD only — Kiro publishes no AUD figure. Its pricing page states prices are "exclusive of applicable taxes and duties, including VAT and applicable sales tax," and Australia is explicitly on the list of supported billing countries — so expect Australian GST to be added through the AWS billing layer at checkout rather than baked into the sticker price. The AUD figures above are our own conversions at roughly 1.40 to the dollar, shown ex-GST. New users get 500 bonus credits valid for 14 days, and your first upgrade to a paid plan earns a US$20 credit. One more wrinkle worth flagging for AWS GovCloud users: pricing there runs about 20% higher and has no free tier.

How it works / features

You write a spec; Kiro builds to it. The pieces that matter:

Specs. The core workflow. Kiro turns a feature request into a requirements document (EARS format), a technical design, then a discrete task list it works through. The spec is a living artefact you can edit, and it keeps the AI honest about what it is meant to be building.
Property-based testing (added at GA). Rather than testing a handful of examples, Kiro extracts "properties" from your spec — universal rules like "for any authenticated user and any active listing, the user can view it" — and throws hundreds of generated cases at your code to find where it breaks. It is closer to a red-team than a unit test.
Agent hooks and steering files. Hooks trigger the agent automatically on events (save a file, run a test). Steering files give the agent persistent project rules and context so you are not repeating yourself every session.
Checkpointing. Roll back any number of steps in an agent run without losing later progress — a credit-saver when a long task goes sideways.
Multi-model AI. Auto (the default agent, cheapest on credits) plus your pick of Claude Opus 5, Opus 4.8, Sonnet 5, Sonnet 4.6, Sonnet 4.5 and Haiku 4.5, OpenAI's GPT-5.6 (Sol, Terra, Luna) and open-weight models. Premium models are paid-tier only and not available in every country.
Kiro CLI. A terminal agent that shares your subscription, credits, steering files and MCP setup with the IDE — handy for automation and CI/CD.
Multi-root workspaces, MCP, and ACP. Work across several project roots at once, connect external tools via MCP, and use the agent from ACP-compatible editors.

On code ownership, Kiro is on the right side of the line: it edits files on your own machine and your code is yours, in your repo. This is a real IDE, not a hosted app builder that keeps your backend — a meaningful difference from the no-code crowd.

What it does well

The spec discipline genuinely produces better output. The single biggest failure mode of AI coding is the agent confidently building the wrong thing. Forcing a requirements-and-design step up front means you catch the misunderstanding on paper, in two minutes, instead of in 200 lines of generated code. For anything beyond a throwaway script, that structure pays for itself.

Property-based testing is a real differentiator. Most AI tools generate tests that flatter the code they just wrote. Kiro's approach — extract rules from the spec, then generate hundreds of cases to break them — is a more honest check on whether the implementation matches intent, and nothing else in the category does it the same way.

It runs the frontier models, and more of them than it used to. Kiro's own FAQ lists Anthropic's Claude Opus 5, Opus 4.8, Sonnet 5, Sonnet 4.6, Sonnet 4.5 and Haiku 4.5; OpenAI's GPT-5.6 in its Sol, Terra and Luna variants; and open-weight models including DeepSeek 3.2, MiniMax M2.5, GLM-5 and Qwen3 Coder Next (checked 21 September 2026). Getting that menu through one subscription, with the Auto agent quietly optimising cost, is a strong hand. The Auto-versus-named-model credit maths even gives you a real lever over spend.

AWS backing brings indemnity and a grown-up security posture. Kiro is built and run by AWS on the same infrastructure standards as the rest of its cloud, and AWS offers IP indemnity for Kiro's output under its service terms — a level of legal cover most AI-IDE startups simply cannot match.

It feels like the editor you already use. Built on Code OSS with Open VSX extension support, Kiro imports your VS Code habits, keybindings and many extensions on day one.

Where it falls short

This is a developer tool, full stop. Kiro is excellent at making developers more disciplined. It will not let a non-technical Melbourne business owner describe an app and walk away with a working product — you still need to read code, run a terminal and reason about a technical design. If that is you, a no-code builder is the right category, not this.

The spec workflow is friction as well as discipline. Writing requirements and a design before any code is the point, but it is also slower and more demanding than just prompting. For quick prototypes and small jobs, the ceremony can feel like overkill, and there is a real learning curve to using specs well.

Credit burn and monthly expiry. Spec tasks, refinement and hook runs all draw from one pool, and named premium models burn faster than Auto. Get into a debugging loop on a complex task and the credits go quickly — and whatever you do not use each month is gone, with no rollover. Budgeting genuine usage is hard, the same structural complaint that runs through this whole category.

Your prompts can train models — unless you pay up. This is the one to read twice. On the Free tier and individual subscriptions, AWS may use your content — questions, inputs, and the generated responses and code — for service improvement, including model training. Only enterprise users (or those accessing Kiro through an Amazon Q Developer Pro subscription) are excluded by default. For an Australian business with sensitive or client code, that is a material consideration on the cheaper tiers.

No Australian data residency. For free and individual users, your content is stored in US East (N. Virginia). Enterprise customers with residency needs can pin storage to N. Virginia or Europe (Frankfurt) — but there is no Australian region at all, and cross-region inference means processing can move within that geography regardless. For Privacy Act-sensitive work, US-default hosting is the deciding issue.

Australian fit

Pricing: US dollars, no AUD shown. Australia is a supported billing country and Kiro's pricing is stated exclusive of sales tax, so expect GST to be added via the AWS billing layer at checkout — confirm it there rather than assume, and budget for the exchange rate on top. Data residency: US East by default for individuals; Frankfurt only for enterprise; no Australian region, and processing can cross regions within the storage geography. Data use: free and individual content may be used for AWS service improvement and model training; only enterprise is excluded by default — a real concern for sensitive Australian code. Support: no Australian entity, presence or local-time support of its own; everything routes through AWS globally, with billing support via the account dashboard. Availability: Australians can buy any tier freely and the product is globally accessible — but there is nothing Australia-specific in it. The AWS ownership helps on stability and indemnity, not on data sovereignty. Australian Fit: 2/5.

Score

Criterion	Score	Why
Value for Money	3/5	Competitive US$20 entry and the Auto-vs-named-model credit lever help, but credit burn on spec tasks, monthly expiry and USD billing make true cost unpredictable.
Ease of Use	3/5	Familiar VS Code base and a guided spec flow, but it is a developer tool with a real learning curve — not usable by non-coders.
Feature Depth	4/5	Specs, property-based testing, hooks, steering, checkpointing, CLI, multi-root, MCP/ACP and top-tier Claude models. Deep and distinctive.
Australian Fit	2/5	No AUD, GST added via AWS billing, US-default hosting with no AU region, free/individual prompts may train models, no AU support.
Reliability & Support	4/5	AWS-built and operated, GA since Nov 2025, IP indemnity under AWS terms — offset by no Australian support hours and the credit-budgeting friction.
Total	16/25

⭐⭐⭐⭐ Very Good (16–20).

Australian Fit is 2/5, below the 4/5 threshold in our methodology, so Kiro cannot be named a category Top Pick regardless of its score — the raw 16/25 still earns an honest four stars.

Bottom line

If you are a developer who is tired of AI agents building the wrong thing, Kiro is one of the most thoughtful tools going. The spec-first workflow is real discipline, property-based testing is a genuine edge, and you get the newest Claude models with AWS reliability and indemnity behind them. For structured, production-minded work, it earns its place.

If you are a non-technical owner who wants an app built for you, this is the wrong category. Kiro makes developers better; it does not replace one. Look at a no-code builder instead and come back to Kiro if you hire someone who codes.

If your code is sensitive or client-confidential, read the data terms before you sign up. On the free and individual tiers, AWS may use your prompts and generated code for model training, and everything is stored in the United States. Only the enterprise tier excludes your content from service improvement by default — so for regulated or commercially sensitive Australian work, price in enterprise or treat the cheaper tiers with caution.

If you are choosing between Kiro and a pure agentic coder like Cursor or Claude Code, Kiro's bet is structure: write the spec, build to it, test the properties. Cursor and Claude Code are faster and looser. Pick on temperament — if you want the AI on rails with a paper trail, Kiro is the better fit; if you want speed and trust your own judgement, the others move quicker.

Update history

9 Oct 2026 — Housekeeping: replaced internal editing references in this history with plain wording. No fact, price or score changed.
29 Sep 2026 — verified USD by browser; ladder clean. kiro.dev/pricing on an Australian connection: Free $0 (50 credits), Pro $20 (1,000), Pro+ $40 (2,000), Pro Max $100 (5,000), Power $200 (10,000), add-on credits $0.04 — every figure holds; bare dollar sign, no A$, prices "exclusive of applicable taxes and duties". New on the page: a US$20 credit toward the first paid month for sign-ups via social login or AWS Builder ID. The model list in the feature bullet and the models FAQ (body and schema) still named Opus 4.5–4.7 and omitted Opus 5, Sonnet 5 and GPT-5.6; both now match the vendor's current FAQ, which the 21 Sep note had already applied to the verdict section. Source · capture Dossiers/captures/kiro-pricing-2026-09-29.txt
21 Sep 2026 — Model-claims check (our sourcing rules), checked against Kiro's own FAQ answer “Which models does Kiro use under the hood?” on 21 September 2026. Still true: Opus 4.8 and Sonnet 4.6 are both still selectable, and the credit multiplier is quoted verbatim from the same FAQ (“a given task that consumes X credits to execute in Auto, will cost you 1.3X credits to execute via Sonnet 4.6”). Now false, corrected: the framing “including the newest Claude” — Kiro now also offers Claude Opus 5, Sonnet 5 and Haiku 4.5, so Opus 4.8 is no longer the newest on its own menu. New and previously unreported: Kiro now also runs OpenAI's GPT-5.6 (Sol, Terra, Luna) and four open-weight models, which the review did not mention at all. Score unchanged at 16/25 — the model menu has widened, not narrowed, and Feature Depth was argued on spec-driven development and property-based testing. reviewed: not moved.
25 Jun 2026 — First reviewed.
24 Sep 2026 — The AUD column was never grossed up for GST, and every word around it said it was. The header read “gross-up; GST unconfirmed” and the note beneath described a 10% uplift — while the figures themselves are plain conversions at 1.4045. Checked before changing a word: across the 63 files carrying this wording, 141 rows convert at 1.4045 and none at 1.4045 × 1.10. No figure changed here; only the description of them did. Two things were wrong beyond the arithmetic: gross-up names a tax uplift and was being used for a currency conversion, and GST unconfirmed is stale — our GST rule settled it on 5 September 2026, and a GST-registered Australian business supplying its ABN is not charged GST on an imported digital service. No USD price moved and no score changed. reviewed: untouched per our dating rule.

Frequently asked questions

Do I need to know how to code to use Kiro?

Yes. Kiro is an AI IDE for developers, not a no-code app builder. It makes coders more disciplined by working from a written spec, but you still need to read code, use a terminal and reason about a technical design. Non-technical users should look at a no-code builder instead.

Is Kiro's pricing in Australian dollars?

No. Pricing is in US dollars — Free, then Pro at US$20/mo, Pro+ at US$40, Pro Max at US$100 and Power at US$200, each with a monthly credit allowance and $0.04/credit overage. Australia is a supported billing country and prices are stated exclusive of sales tax, so expect GST to be added at checkout via AWS billing.

Does Kiro use my code to train AI models?

On the Free tier and individual subscriptions, yes — AWS may use your prompts, inputs and generated code for service improvement, including model training. Enterprise users (and those accessing Kiro through an Amazon Q Developer Pro subscription) are excluded by default. For sensitive code, that points to the enterprise tier.

Where is my data stored with Kiro?

For free and individual users, in the United States (US East, N. Virginia). Enterprise customers with residency needs can choose N. Virginia or Europe (Frankfurt). There is no Australian region, and cross-region inference means processing can move within the storage geography.

What AI models does Kiro run?

By default the Auto agent picks a mix of frontier and specialist models for each task. You can also pick a specific model on paid tiers: Claude Opus 5, Opus 4.8, Sonnet 5, Sonnet 4.6, Sonnet 4.5 or Haiku 4.5, OpenAI's GPT-5.6 (Sol, Terra, Luna), or open-weight models such as DeepSeek 3.2 and Qwen3 Coder Next. The free tier includes Claude Sonnet 4.5 and open-weight models, rate-limited.

Is Kiro generally available or still in preview?

Generally available. Kiro launched in preview in July 2025 with a waitlist and reached general availability on 17 November 2025 — the waitlist is gone, and GA added property-based testing, checkpointing, a terminal CLI and team plans.

Sources

https://kiro.dev/pricing/ (primary; tiers, credits, overage, supported countries, tax statement)
https://kiro.dev/faq/ (primary; credit definition, models, rollover, GovCloud) — re-read 21 September 2026 for the model line-up and the 1.3× Sonnet 4.6 credit multiplier — Wayback 21 September 2026
https://kiro.dev/blog/new-pricing-plans-and-auto/ (primary; credit model, Auto agent, fractional credits, 15 Sep 2025)
https://kiro.dev/blog/general-availability/ (primary; GA 17 Nov 2025, property-based testing, CLI, team plans, checkpointing)
https://kiro.dev/docs/privacy-and-security/data-protection/ (primary; US East storage, EU Frankfurt for enterprise, content used for model training on free/individual tiers)
https://kiro.dev/ (primary; spec-driven positioning, "beyond AI coding to agentic engineering")
https://kiro.dev/docs/guides/migrating-from-vscode/ (primary; Code OSS base, Open VSX)
https://aws.amazon.com/blogs/aws/aws-weekly-roundup-kiro-waitlist-ebs-volume-clones-ec2-capacity-manager-and-more-october-20-2025/ (secondary; preview/waitlist history, 100k developers)
https://visualstudiomagazine.com/articles/2025/07/21/forked-again-awss-kiro-latest-ai-assistant-based-on-vs-code.aspx (secondary; July 2025 launch, VS Code fork)

Wayback Machine snapshots queued.

Kiro Review (2026): AWS's Spec-Driven AI IDE, Tested for Australia