Agent Definition Language (ADL): An Open Standard for Defining AI Agents (github.com)

The introduction of the Agent Definition Language (ADL) marks a significant step towards standardizing the definition of AI agents across various platforms. This vendor-neutral, open standard allows for the consistent and interoperable description of an agent's identity, capabilities, configuration of large language models (LLMs), permissions, and dependencies. By providing a shared language that details what an agent is and what it can do, ADL addresses several systemic challenges faced by enterprises, such as inconsistent tool contracts and the difficulties of agent portability across different systems. ADL not only enhances predictability and auditability but also ensures that agents are created with explicit governance metadata, which is crucial for compliance. Unlike other definitions that focus on execution or communication protocols, ADL is dedicated to establishing a clear and coherent framework that models agent competencies without getting into the intricacies of app-level infrastructure. Open-sourced under the Apache 2.0 license, ADL aims to foster community contributions and ensure it transforms into a widely accepted standard, paving the way for faster and more efficient adoption of AI agents in various industries.