Agent Definition Language: The open standard AI agents have been missing

How ADL brings OpenAPI-style standardization to AI agent development.

Open source has always advanced by establishing shared languages. When APIs needed consistency, the community created OpenAPI. When infrastructure required reproducibility, Terraform and Kubernetes YAML emerged. When containers needed interoperability, Open Container Initiative (OCI) unified the ecosystem.

The problem: Why AI agent definitions are fragmented

AI agents are now reaching a similar turning point. They are rapidly becoming core components in modern systems—reasoning across data, orchestrating workflows, and coordinating multi-step logic. Yet unlike APIs, containers, or infrastructure, AI agents lack a common definition layer. Every framework describes agents differently, and the result is fragmentation, duplicated effort, governance gaps, and limited interoperability.

The Agent Definition Language (ADL) introduces a declarative, framework-agnostic specification for describing AI agents in a structured, portable, and transparent way. This article outlines why ADL exists, the problem it addresses, and why it matters to the All Things Open community.

Today, the behavior of an AI agent is often hidden inside:

• Scattered prompts  
• Python classes or wrappers
• JSON fragments
• Framework-specific configuration structures
• Assumptions not visible to reviewers

This makes it difficult for teams to answer basic questions such as:

• What can this agent do?
• What tools is it allowed to call?
• What inputs and outputs does it expect?
• What rules or constraints govern its behavior?
• Who approved this agent for deployment?

The lack of a declarative definition layer leads to non-portable agents, inconsistent governance, gaps in security review, and collapsed visibility across engineering, ML, and compliance teams.

Read more: AI vs ML vs DL: A practical guide with real-world engineering examples

What the Agent Definition Language provides for developers

ADL consolidates the full definition of an agent into a single structured document that includes:

• Role and purpose  
• Capabilities and inputs  
• Tool and permission boundaries  
• Workflow logic  
• Safety and governance constraints  
• Metadata and authorship  
• Versioning and lifecycle information 

By standardizing these components, ADL makes agents:

• Inspectable  
• Reviewable  
• Shareable  
• Testable  
• Governable  
• Portable across frameworks and runtimes  

In the same way that OpenAPI enables consistent API contracts, ADL aims to unify how teams describe and collaborate on agents.

How ADL relates to existing technologies

ADL is intentionally complementary to the systems shaping today’s AI engineering landscape:

A2A (Agent-to-Agent Protocols)

• Define how agents communicate and pass messages
• ADL defines what an agent is—not how it communicates  

MCP (Model Context Protocol)

• Standardizes how tools and context are delivered to models 
• ADL standardizes agent identity, capabilities, and governance  

OpenAPI

• Declarative API contracts that improved interoperability across systems  
• ADL applies similar principles to agents

Instructor (Structured output library)

• Ensures model responses adhere to defined schemas at inference time  
• ADL defines overall agent structure, workflow, and constraints at design time 

These systems operate at different layers. ADL fills the definition layer that none of them currently address.

Why ADL matters for open source AI development

The All Things Open community is grounded in openness, transparency, collaboration, and standards-driven innovation. ADL reflects these values by enabling:

• Interoperability – Agents defined once can run across multiple runtimes  
• Transparency –  Capabilities, tools, and workflows are explicit and reviewable  
• Governance –  Safety, permissions, and policies are clearly encoded  
• Reusability –  Agents can be shared, forked, improved, and versioned  
• Ecosystem growth –  Editors, validators, registries, and testing tools can emerge around a stable specification  

For practitioners building production AI systems, ADL helps move agents out of ad-hoc implementations and into well-defined, auditable components.

The future of the Agent Definition Language standard

ADL is intentionally early. Successful open standards emerge through iteration, critique, real-world use, and community participation. The goal is to evolve ADL with input from developers, researchers, framework maintainers, and open-source contributors.

There is an opportunity for the community to help shape what the agent ecosystem becomes not through proprietary formats, but through open, extensible, vendor-neutral definitions.

How to contribute to the Agent Definition Language

If you believe open standards strengthen the ecosystem, please support the ADL initiative:

• Star the GitHub repo to show interest  
• Fork the repository to experiment or extend the specification  
• Clone the repo to explore ADL locally and provide feedback  
• Open issues or discussions to guide the evolution of the standard 

GitHub Repo: https://github.com/nextmoca/adl

With the help of the All Things Open community, we can build a shared, open, and durable foundation for defining the next generation of AI agents.

More from We Love Open Source

• Deep dive into the Model Context Protocol
• What is artificial intelligence (AI) and the three things it does well
• AI vs ML vs DL: A practical guide with real-world engineering examples
• What is machine learning and how does AI actually learn?
• What is deep learning and how does it work?