This article was originally published on aicoderscope.com
Amazon Q Developer IDE is dead. New signups closed on May 15, 2026 — five days ago. Full end-of-support for paid subscriptions lands April 30, 2027. AWS is putting everything behind its replacement: Kiro, an agentic IDE that launched preview in July 2025, attracted 250,000 developers, and makes a bet so specific it will either look visionary or foolish in 12 months.
The bet: developers generate too much code before they've thought the problem through. Kiro forces you to write a spec — requirements, design, tasks — before touching a line of implementation. Only after you've reviewed and approved the plan does the agent start coding.
That workflow is genuinely different from every other AI IDE on the market. Whether it's different in a way that's worth $20, $40, or $200 a month depends entirely on what kind of code you're shipping.
This review covers verified pricing, the credit model, the actual workflow, where it beats Cursor, and where it plainly doesn't.
What Q Developer shutdown means for you right now
If you've been on Q Developer Pro, you can keep your subscription and keep adding users until April 30, 2027. No forced migration yet. But AWS has drawn a clear line: the latest coding models — including Claude Opus 4.7, which arrived on Kiro May 7, 2026 — are Kiro-exclusive. Q Developer Pro still has Opus 4.5 and existing models. It won't get Opus 4.7 or anything newer.
AWS console Q Developer (the chat assistant inside the AWS Management Console), Slack/Teams integrations, and the marketing/documentation site assistants are not affected by this sunset. Those continue.
The IDE plugin you've been using in VS Code or JetBrains is what's going away. If you want AWS's newest models with agentic coding, Kiro is the only path.
The spec-driven premise
Kiro's core insight is that most AI coding mistakes aren't coding mistakes — they're architectural mistakes that happen before a line of code is written, then get discovered at 400 lines in when an entire feature needs to be torn up.
Spec-driven development runs through three sequential artifacts:
requirements.md captures what you're building using EARS-style notation (Easy Approach to Requirements Syntax): user stories, acceptance criteria, edge cases. Kiro generates a draft from your prompt; you review and edit it.
design.md covers technical architecture: system components, data flow, API contracts, database schema, and integration points. Again, Kiro drafts it; you approve or revise.
tasks.md breaks the design into sequenced implementation steps with explicit dependencies. Each task maps to specific files and functions. This is what the agent actually executes.
The idea is that design mistakes cost minutes to fix at the requirements stage and days to fix after implementation. On a feature that touches five tables, three API endpoints, and two frontend components, this is obviously true. On a feature that changes a button label, it is obviously overkill.
Kiro acknowledges this with two entry points: Feature Specs for the full three-phase workflow, and a Bugfix Spec that uses a compressed diagnosis-then-fix variant. For quick questions and small edits, Vibe Mode skips specs entirely — it's a standard AI chat interface, conceptually identical to Cursor Chat.
Pricing: the credit model
As of March 19, 2026, Kiro's pricing:
| Tier | Price | Credits/month | Effective cost/credit |
|---|---|---|---|
| Free | $0 | 50 (+ 500 bonus first 30 days) | — |
| Pro | $20/mo | 1,000 | $0.020 |
| Pro+ | $40/mo | 2,000 | $0.020 |
| Power | $200/mo | 10,000 | $0.020 |
| Overage | Pay-per-use | — | $0.04/credit |
Two request types consume credits at different rates:
- Vibe requests (chat, quick edits): $0.04/credit = one request per credit
- Spec requests (multi-phase spec generation and execution): $0.20/credit = five times the cost
Kiro's "Auto" mode — the default in the chat window — routes tasks across multiple models including Claude Sonnet 4 and specialized routing models. Auto consumes 1.3× the credits of running Sonnet 4 directly on the same task.
The math on Pro: 1,000 credits works out to roughly 1,000 Vibe-mode requests, 200 Spec-mode requests, or a realistic mix of ~700 Vibe and ~150 Spec per month. Compare that to Q Developer Pro at $19/month, which gave roughly 1,000 standard requests with no credit-splitting complexity. If you were a heavy Q Developer Pro user, Kiro Pro at $20 delivers meaningfully less volume, with the billing complexity of monitoring two different request types.
First-time upgraders get a one-time $20 credit applied to their subscription. New signups also receive 500 bonus credits valid for 30 days, giving you a genuine two-week window to evaluate Spec mode before paying.
AWS Startups members can get up to one year of Kiro Pro+ complimentary through the AWS Startups network — worth verifying on the Kiro pricing page if you qualify.
Models available
Paid tiers (Pro, Pro+, Power) get access to:
- Claude Opus 4.7 with adaptive thinking (added May 7, 2026) — Kiro-exclusive, not on Q Developer
- Claude Opus 4.6 and Claude Sonnet 4.6 — both with 1M context windows
- Claude Sonnet 4 and Claude Haiku
- GLM-5 — a sparse MoE model with 200K context, useful for repository-scale context
Auto mode blends these behind an intent-detection layer, selecting models based on task complexity. You can override and specify a model directly in settings.
Agent Hooks: automation that actually runs
Agent Hooks are the feature that most reviews undersell. They let you define actions that fire automatically when IDE events occur:
- File events: file created, saved, deleted
- Lifecycle events: before/after a prompt is submitted, before/after a tool runs, when the agent stops
- Spec events: before/after a spec task executes
- Manual triggers: run a hook on demand
Each hook can run either a predefined agent prompt or a shell command. Practically, this means you can configure Kiro to auto-generate unit tests when a file is saved, run a security scan when a new file is created, or validate API contracts before a spec task completes — all without manual invocation.
This is a real workflow capability, not a marketing feature. Cursor doesn't have an equivalent built-in hook system. Building this kind of automation with Cursor requires external scripts and task runners.
The catch: Agent Hooks are most useful when you've already committed to Kiro's workflow. Teams that use Kiro as an occasional overlay on top of Cursor won't bother configuring hooks. The payoff scales with commitment.
Autopilot vs Supervised mode
Kiro offers two execution modes for all agent tasks:
Autopilot — the agent makes changes across your codebase, executes tool calls, and completes tasks with minimal interruption. You review the result when it's done.
Supervised — the agent pauses for your approval after each turn that includes file edits. You see exactly what it's about to do before it does it.
For complex multi-file features in production codebases, Supervised is the sane default. Autopilot is best on greenfield features or contained modules where unexpected file edits have limited blast radius.
Kiro vs Cursor: the honest comparison
These two tools represent different theories about where AI development failures happen.
| Kiro Pro | Cursor Pro | |
|---|---|---|
| Price | $20/mo | $20/mo |
| Credits/requests | 1,000 credits (~700 vibe + ~150 spec) | Unlimited standard AI usage |
| Tab autocomplete | Basic | Supermaven model, ~72% acceptance rate, sub-200ms |
| Agentic approach | Spec-first: plan → review → implement | Prompt-first: immediate multi-file Agent mode |
| Structured planning | Built-in (requirements.md / design.md / tasks.md) | Plan mode (less enforced, no artifact output) |
| Agent hooks | Native (file events, life |
Top comments (0)