Kiro — Three Months from Custom Subagents to CLI 2.0

Why This Matters

1. Spec-driven development is the wedge. The arc starts with new spec types (Feb 18) and ends with "specs just got faster (and smarter)" + requirements-bug detection (both May 12). The whole quarter is Kiro deepening its spec-first identity rather than competing feature-for-feature on chat-based coding.
2. CLI graduated into a peer product. February's CLI was a terminal companion; April's CLI 2.0 ships Windows support, headless mode, and programmatic invocation — i.e., it's now a real CI/CD primitive.
3. Enterprise surface keeps expanding — government cloud, identity + usage metrics, MCP and model governance, granular code review.
4. Model breadth as a feature — Opus 4.6 → Opus 4.7, Sonnet 4.6, open-weight models, MiniMax M2.5, GLM-5. Kiro is not betting on any single provider.

The Arc, In Five Phases

Phase 1 — Early February (the v0.9 cluster, Feb 5)

• Kiro 0.9: Custom subagents in the IDE, new enterprise controls, and granular code review — the headline release.
• Specialized IDEs deserve AI too: Kiro adopts ACP — protocol support for non-VSCode IDEs.
• Refactoring made right: how program analysis makes AI agents safe and reliable — the technical pitch behind why Kiro's edits are different.
• Opus 4.6 is now available in Kiro — model freshness on launch day.

Phase 2 — Late February (enterprise + spec types, Feb 13–27)

• Enterprise identity and usage metrics (Feb 13)
• Claude Sonnet 4.6 is now available in Kiro (Feb 17)
• Kiro loves regulated workloads — government cloud support (Feb 18)
• New spec types: fix bugs and build on top of existing apps (Feb 18) — this is the load-bearing post. Spec types now include "fix this bug" and "extend this app," not just "build something new."
• The bug fix paradox: why AI agents keep breaking working code (Feb 19) — Kiro is publicly arguing that the conventional agent loop is structurally bad at bug fixes, and positioning specs as the alternative.
• The hidden inefficiencies in AI coding (and how we find them) (Feb 23)
• Surgical precision with AST-based code editing in Kiro (Feb 27) — the technical foundation: AST edits, not text patches.

Phase 3 — March (governance, identity tiers, the AAAI talk)

• Enterprise governance: control your MCP servers and models (Mar 12) — admins can constrain which MCP servers and which models a team can use.
• Introducing the Kiro Students tier (Mar 18)
• A new look for the Kiro CLI (Mar 23) — UI refresh setting up CLI 2.0.
• From copilots to coworkers at AAAI: the gap between agentic research and production (Mar 24) — Kiro showing up at AAAI is itself a signal about how the team wants to be positioned.
• Bringing agentic AI to silicon development (Mar 26) — a vertical push into hardware/EDA workflows.

Phase 4 — April (the CLI 2.0 inflection)

• MiniMax M2.5 and GLM-5 are now in Kiro (Apr 2)
• Bringing back the Kiro startup credits program (Apr 7)
• Planview saves 40+ hours per audit cycle by automating SOC 2 compliance with Kiro CLI (Apr 10) — the kind of enterprise case study that signals CLI is the surface enterprises are actually adopting.
• Kiro CLI 2.0: a new look and feel, headless CI/CD pipelines, and Windows support (Apr 13) — the major release of the window. Windows support closes the last platform gap; headless CI/CD turns Kiro into a build-pipeline primitive.
• Run Kiro CLI programmatically: introducing headless mode (Apr 13) — companion post detailing the automation API.
• Root cause in 33 seconds: How Kiro CLI saved 4 years of build time (Apr 16) — the post-launch case study.
• Opus 4.7 is now available in Kiro (Apr 17)
• Build with Kiro: Introducing the community hub and Kiro Labs (Apr 23) — community platform + an experimental-features lab.

Phase 5 — May (community, paid-tier push, spec polish)

• Introducing Kiro Ambassadors (May 7) — formal community program.
• More room to explore: $20 paid tier sign-up bonus (May 8) — paid-tier conversion push.
• Requirements analysis: catching requirement bugs before they become code (May 12) — the spec story keeps maturing. Bugs in the requirements caught before they reach code is the canonical sales pitch for spec-driven development.
• Specs just got faster (and smarter) (May 12) — performance + intelligence improvements to the spec engine.

What The Shape Tells You

1. Kiro is not trying to win the chat-based coding race. The blog goes out of its way to argue that chat-driven, text-patch-based agents are structurally bad at certain classes of work (especially bug fixes and refactoring of existing code). Specs + AST edits + program analysis is the alternative pitch, and the team is investing in it at the product, marketing, and research levels.
2. The CLI is the enterprise wedge. IDE adoption is a slow individual-developer motion; the CLI lands in CI/CD pipelines and audit workflows where the buying decision is centralized. The Planview SOC 2 case study (Apr 10) and CLI 2.0 (Apr 13) are the same play.
3. Multi-model is a feature, not a hedge. Opus 4.6 → 4.7, Sonnet 4.6, open weights, MiniMax M2.5, GLM-5 — Kiro is explicitly avoiding being a single-provider product. That mirrors what we wrote up about Perplexity Computer — model-orchestration as the meta-layer that doesn't depend on any one lab winning.
4. Spec-driven development is the bet. If you only read four posts in the window, read "New spec types" (Feb 18), "The bug fix paradox" (Feb 19), "AST-based code editing" (Feb 27), and "Requirements analysis" (May 12). Together they describe a coherent product philosophy that the rest of the AI-coding field doesn't have.

Our Take