Agentic AI
AWS Kiro embraces open protocol to let developers swap coding agents freely
By standardizing how clients talk to AI agents through a shared protocol, developers may soon choose their editor and AI assistant independently.
The New Stack
AWS has restructured its Kiro coding agent to decouple it from specific editors and terminals, moving developers closer to a world where they can pick their development tools and AI assistants separately. The shift hinges on adopting the Agent Client Protocol (ACP)—an emerging standard developed jointly by Zed and JetBrains—as the interface between Kiro's agent and its client applications.
From fragmented harnesses to a unified boundary#
Previously, Kiro's agent operated through three separate language-specific harnesses: TypeScript for the IDE, Rust for the command line, and Python for the web. AWS has now consolidated these into a single agent process that communicates exclusively through ACP, treating the agent-client relationship as a standardized boundary rather than a proprietary connection.
This architectural move solves a persistent problem in extensible systems. Earlier attempts using shared libraries allowed clients to accumulate agent-specific logic by tapping into internal APIs, eroding the separation of concerns. Running the agent as a standalone process and forcing all communication through a standard protocol maintains clean boundaries regardless of whether the agent executes locally or in a cloud sandbox.
Standardization without losing differentiation#
Rather than creating a proprietary variant, AWS extended ACP itself—introducing more than 20 agent-callable methods, 15 client-callable methods, and 20 notification types under the _kiro/ namespace. The company also added WebSocket transport for web and iOS clients while preserving ACP's stdio transport for local deployment. Each client can now optimize its user experience and platform-specific features while sharing a common configuration model.
The pattern mirrors what the Language Server Protocol achieved for language tooling: instead of requiring every editor to build custom integrations for every agent, both sides target one common protocol. This reduces the complexity of N × M one-off connections to N + M standardized ones.
Governance as the new competitive battleground#
As communication becomes commoditized through protocols, AWS is positioning governance as its primary differentiation layer. Kiro unifies two incompatible permission systems—regex-based CLI rules and prefix-matching IDE rules—into a capability model backed by , AWS's authorization language. Capabilities like , , , , , and represent functional classes of operations rather than individual tool approvals.

