AWS Kiro Adds $100/Month Pro Max Tier and Self-Correcting Subagents — Amazon Q Developer Sunsets April 2027
AWS’s Kiro IDE is moving fast on pricing and agentic features. Three updates shipped this week: a new Pro Max subscription tier, a self-correcting multi-agent pipeline, and GitLab support in Kiro Web. Combined with Claude Opus 4.8 availability added last week, the product is now positioned as a full-stack competitor to Cursor, Claude Code, and Google Antigravity.
Pro Max: Closing the Gap Between $40 and $200
Kiro’s previous tiers jumped from $40/month (Pro+, 2,000 credits) to $200/month (Power, 10,000 credits) with nothing in between. Pro+ users consistently reported blowing past allocations and paying unpredictable overage fees to stay productive.
Pro Max lands at $100/month with 5,000 credits and access to all premium models — Claude Opus 4.8, Sonnet 4.6, and the Auto router. It carries the same feature set as the Power plan: Specs, custom subagents, Powers, Hooks, and full CLI access. The key delta over Power is the credit ceiling: 5,000 vs 10,000 per month, with per-credit overage at $0.04 for teams that exceed it.
The full pricing stack:
| Tier | Price | Credits/Month | Models |
|---|---|---|---|
| Free | $0 | 50 | Sonnet 4.5 + open-weight |
| Pro | $20 | 1,000 | Premium |
| Pro+ | $40 | 2,000 | Premium |
| Pro Max | $100 | 5,000 | Premium (all) |
| Power | $200 | 10,000 | Premium (all) |
Self-Correcting Subagent Pipelines
The more technically interesting release is the self-correcting subagent pipeline. When a coding task is dispatched to an implementer subagent, a reviewer stage can now return the work with specific failures before it surfaces to the user. The implementer loops until the reviewer passes it. The system handles code review and refactoring autonomously rather than surfacing every intermediate failure state as a user decision.
This is a meaningful shift from the baseline multi-agent pattern where human review gates every agent output. For large refactoring tasks or spec-driven feature builds, having the review loop run inside the agent workflow cuts the number of required human interventions per task.
Amazon Q Developer: Official Countdown
AWS is not managing two agentic coding products indefinitely. The deprecation calendar is now fixed:
- May 15, 2026: New Q Developer signups blocked. Existing subscriptions can add users but cannot scale new accounts.
- May 29, 2026: Claude Opus 4.6 and later models removed from Q Developer. All frontier-tier models are now Kiro-exclusive.
- April 30, 2027: Full end of support for Q Developer IDE plugins and paid subscriptions. Code transformation features become inaccessible.
Any team still running Q Developer Pro is now capped at Claude Opus 4.5 for AI-assisted coding. The practical deadline to access frontier models was May 29. The deadline to avoid service disruption is April 2027, but enterprise procurement and team retraining make a last-minute migration risky.
The Spec-Driven Difference
Kiro’s architectural bet is that agentic coding tools fail because intent stays ambiguous. Where Cursor or Claude Code work in a prompt-response loop, Kiro forces three artifacts before code generation begins: requirements.md (user stories and acceptance criteria), design.md (architecture, data models, component design), and tasks.md (sequenced implementation steps). The agent does not write code until a developer approves all three.
Agent Hooks extend this pattern at runtime. Lifecycle-aware automations fire when files are saved, PRs are opened, or specs change — running test updates, documentation generation, or security scans without manual triggers. For teams already running CI/CD pipelines, the paradigm is recognisable.
Kiro routes across Claude Sonnet 4.5, Opus 4.8, Amazon Nova Pro, and an Auto mode that balances model capability against credit cost. It integrates natively with AWS: IAM-scoped least-privilege policy generation, CDK and CloudFormation generation, AWS MCP Server support (GA since May), and CloudTrail audit logging.
Competitive Context
The agentic IDE market has three credible players at the frontier tier right now. Cursor holds the largest developer install base and ships Composer 2.5 with its own coding model. Claude Code dominates at startups where Anthropic’s direct API access matters. Google Antigravity (GA June 11, with Gemini CLI shutting June 18) is the entrant most directly competing with Kiro for enterprise and cloud-native teams. Kiro’s differentiation is AWS-native depth and the spec-driven discipline — a deliberate constraint that positions it for engineering teams already running structured software processes rather than exploratory vibe-coding workflows.
Kiro Web (browser-native, no desktop install) now supports GitLab with full merge-request automation, closing the gap for teams running private or self-hosted GitLab instances. GitHub was already supported.
The Pro Max tier is available immediately from account settings, with mid-cycle proration. New accounts can apply the $20 sign-up credit toward the first Pro Max billing cycle.