AWS Open-Sources Dogwood to Set Rules for AI Agent Tool Calls

Amazon Web Services has open-sourced Dogwood, a project aimed at establishing rules for AI agent tool calling, according to a report from InfoQ China. The project is positioned as an open-source effort to govern how agents invoke external tools.

As large models move from conversation to autonomous execution, agents increasingly rely on external tools — APIs, databases, web services, and enterprise systems — to complete real tasks. Whether those tool calls are well-governed and controllable has become a deciding factor in agent reliability.

Dogwood is positioned precisely at this point: an open-source effort to give tool invocation a set of rules and constraints so agent behavior becomes more predictable. InfoQ reports that the project is being open-sourced by AWS, with technical implementation details expected to be published in the project documentation.

The move reflects the broader agent-engineering wave. Over the past year, vendors have competed fiercely on model capability; as models grow more comparable, engineering concerns such as tool calling, orchestration, permissions, and observability are becoming the new differentiator.

Open-sourcing means developers can inspect, adopt, and contribute to the tool-calling specification, giving the industry a shared foundation for standardization. For enterprise users, an open rule layer could lower the integration and governance cost of deploying agents.

It is also a signal about cloud vendors' changing role in the agent ecosystem: AWS is no longer just a provider of compute and model access, but is seeking influence at the tool layer. The move echoes wider community discussion about standardizing how agents use tools.

What to watch next: the technical details of Dogwood — how it defines tool-calling rules, which runtimes it supports, and whether it integrates with AWS's existing agent services — should become clearer through documentation and community feedback.