Free agents: How AWS Kiro could untie agents from editors
Choosing a coding agent may soon no longer mean having to pick a new editor or terminal, too. That scenario The post Free agents: How AWS Kiro could untie agents from editors appeared first on The New Stack .
AWS is transforming the way its Kiro coding agent interacts with various client applications. Previously, Kiro relied on three distinct agent harnesses - one for its TypeScript IDE, another for the command-line interface (CLI) written in Rust, and a third for the web experience built with Python. This fragmentation made it challenging for developers to select their preferred coding tools and AI agents independently.
AWS has now consolidated these three harnesses into a single standalone agent harness that runs as a separate process alongside the workspace. This simplification streamlines the architecture and makes the system more maintainable.
What's more significant is the architectural decision AWS made by adopting the Agent Client Protocol (ACP) as the interface between Kiro’s clients and its own agent. ACP originated at Zed and is now jointly developed with JetBrains. By treating the client-agent boundary as a standardized interface instead of a proprietary implementation detail, AWS is enabling greater interoperability between different client applications and agents.
The Agent Client Protocol allows clients to communicate exclusively through a shared protocol, while the execution environment (i.e., whether the agent runs locally or inside a cloud sandbox) becomes an implementation detail. This separation of concerns simplifies integration and allows clients to focus on delivering native user experiences tailored to their specific platforms.
AWS extended the Agent Client Protocol by introducing over 20 agent-callable methods, 15 client-callable methods, and 20 notification types, all namespaced under "_kiro/". They implemented a WebSocket transport for web and iOS clients while maintaining the standard stdio transport for local execution. This approach enables third-party developers to create custom agents and lifecycle hooks that can be seamlessly integrated into Kiro's ecosystem.
This architectural shift closely resembles the success of the Language Server Protocol (LSP) in the programming language tooling space. LSP separated editors from language intelligence, enabling a wide range of editors to adopt language-specific intelligence without the need for bespoke integrations.
By standardizing the client-agent boundary, AWS is reducing integration complexity from an N x M relationship (where N is the number of clients and M is the number of agents) toward an N + M model. This approach allows vendors to offer differentiated capabilities while still maintaining interoperability through the standardized protocol.
The most significant differentiation for Kiro lies in its policy management system, which replaces two incompatible permission systems with a unified capability model backed by AWS's Cedar authorization language. This model abstracts permissions around functional intent, allowing for more fine-grained control over what actions agents can perform.
As enterprise adoption of Kiro grows, the governance layer built on top of the standard protocol is likely to become a key differentiator. Features such as identity integration, auditing, authorization, policy inheritance, and sandboxing will be crucial for securing and managing the use of AI agents within organizations.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
