What is the proposed solution for AI agent fragmentation?

The key unlock is standardization around agent-to-agent communication. This involves creating common protocols that allow agents from different frameworks to interact seamlessly.

What is BAI and how does it help with agent interoperability?

BAI (an open-source platform) allows users to discover, run, and compose AI agents from any framework. It aims to facilitate the use of diverse agents and helps in developing the standardization protocol.

How can agents from different frameworks (like custom solutions and LangGraph) work together?

Through standardized communication protocols like ACP, agents built on different frameworks can interconnect. This allows for interchangeable agents and the ability to run them with the same commands and parameters.

How are new agents integrated into the BAI platform?

New agents can be integrated using a 'provider' concept, which acts as a source for templates. These templates can be instantiated as agents, allowing for easy import from sources like public GitHub repositories.

What is the difference between agent frameworks and function calls?

In function calls, the LLM decides which tool to use. Agent frameworks, especially with protocols like ACP, focus on enabling agents to communicate and delegate tasks to each other, allowing for bidirectional communication and complex orchestration.

How does agent discoverability compare to microservices discoverability?

While microservices are generally static, AI agents can be dynamic, even creating other agents on the fly. This dynamism presents new challenges for discoverability and resource management compared to the more established practices in microservices.

How is telemetry and tracking handled in complex agent systems?

Integrated solutions like Phoenix from Arise, which works with the BAI platform, provide open inference capabilities for tracking and tracing agent interactions, helping to manage complex, distributed agent systems.

AI Dev 25 | Kate Blair & Ismael Faro: The future of agent interoperability

DeepLearning.AI

Entertainment4 min read31 min video

Mar 27, 2025|858 views|10

Save to Pod

Key Moments

TL;DR

IBM Research proposes an open standard for AI agent interoperability, enabling seamless communication and workflow automation across platforms.

Key Insights

The AI agent landscape is rapidly fragmenting, with current frameworks being largely incompatible, hindering collaboration and reuse.

A key unlock for AI agents is standardization around agent-to-agent communication, similar to MCP for large language models.

IBM Research's approach involves leveraging existing protocols, the open-source ecosystem, and a feature-driven methodology.

BAI is an open-source platform designed for discovering, running, and composing agents from any framework.

The proposed Agent Communication Protocol (ACP) extends MCP to include agent capabilities beyond prompts, resources, and tools.

Key development areas include discoverability, interoperability between different agent frameworks, and dynamic agent composition for complex workflows.

THE GROWING CHALLENGE OF AI AGENT FRAGMENTATION

The AI agent landscape is experiencing rapid development but is becoming increasingly fragmented. Developers are presented with numerous frameworks, each with its own abstractions and dependencies, making it difficult to switch between them or integrate agents from different sources. This fragmentation leads to significant rework, especially within large enterprises where hundreds of agents might be in production, managed by different teams. The core problem is the lack of a standardized way for these atomic agents to communicate and collaborate effectively.

THE NEED FOR STANDARDIZATION AND INTEROPERABILITY

The primary solution proposed to overcome AI agent fragmentation is standardization around agent-to-agent communication. This concept is seen as the next major unlock for AI, akin to the impact of the Model Context Protocol (MCP) for large language models. The goal is to enable easy discovery and trial of agents from various frameworks without deep technical overhead for each. This standardization would allow for automated complex workflows by composing specialized agents and facilitate swapping out agents for better versions without rewriting entire systems.

IBM RESEARCH'S APPROACH TO INTEROPERABILITY

IBM Research advocates for an open-source community effort to build this standard. Their approach involves standing on the shoulders of giants by looking to existing communication protocols and leveraging the current ecosystem, such as MCP and emerging protocols like NI. Most importantly, they emphasize a feature-driven approach, focusing on real-world jobs-to-be-done to inform the standardization process. This means running numerous agents from different frameworks to identify common needs and build standards from practical use cases.

INTRODUCING BAI AND THE AGENT COMMUNICATION PROTOCOL (ACP)

IBM has developed BAI, an open-source platform designed for discovering, running, and composing agents from any framework. BAI serves as a vehicle for this initiative. Building upon MCP, they have bootstrapped a pre-alpha version of the Agent Communication Protocol (ACP). ACP extends MCP by enabling servers to offer agent capabilities alongside prompts, resources, and tools, allowing for more complex agent interactions and compositions.

DEMONSTRATING DISCOVERABILITY AND FRAMEWORK INTEROPERABILITY

The presented demos highlight crucial aspects of agent interoperability. The first demo showcases how agents can discover each other and how metadata attached to agents facilitates this discovery. It illustrates that agents built with entirely different frameworks, such as custom solutions and LangGraph, can run using the same command line and parameters, thanks to ACP. This emphasizes the ability to swap agents faster and select the best one based on specific needs, a significant step beyond simple function calls.

ENABLING AGENT COMPOSITION AND FLEXIBLE DEPLOYMENT

Further demonstrations explore the composition of agents into complex workflows. Agents can act as supervisors or orchestrators, sequentially chaining other agents and passing outputs automatically. This allows for the creation of intricate processes where one agent can discover and invoke others on the platform. The system also supports dynamic agent integration, where new agents, even from newly emerging frameworks like OpenAI's SDK, can be integrated with minimal code, showcasing flexibility and adaptability.

ADDRESSING CHALLENGES AND FUTURE DIRECTIONS

Discussions surrounding ACP delve into critical challenges such as security, efficient scaling, and dynamic agent creation. Unlike static microservices, agents can be highly dynamic, even creating new agents or tools on the fly, which raises questions about resource management and deployment. The protocol aims to define how computation can be distributed efficiently between agents, moving beyond simple client-server or sequential paradigms. The goal is to foster bidirectional communication and enable sophisticated agentic patterns, rather than just registering agents as tools.

INTEGRATING TELEMETRY AND OPEN COMMUNITY COLLABORATION

The development actively incorporates telemetry and tracking mechanisms to monitor agent interactions. By using an entry point within the communication protocol, the system can track agent activities effectively. This effort aligns with community discussions on security and scaling, exploring solutions like Kubernetes alternatives for agent deployment. IBM Research emphasizes that this is an ongoing, open community effort, inviting developers to visit their webpages to follow discussions and contribute to the advancement of AI agent interoperability.

Mentioned in This Episode

●Software & Apps

●Companies

●Organizations

●Concepts

●People Referenced

Best Practices for AI Agent Interoperability

Practical takeaways from this episode

Do This

Standardize agent-to-agent communication protocols (like ACP).

Leverage existing protocols and ecosystems (e.g., MCP).

Adopt a feature-driven approach, building standards from practical needs.

Use platforms like BAI to discover, run, and compose agents from various frameworks.

Encapsulate agent logic with wrappers to enable standardization.

Consider using supervisor agents to orchestrate complex workflows and compose agents sequentially.

Implement telemetry and tracing for better observability in distributed agent systems.

Avoid This

Do not assume agents are static; prepare for dynamic creation and evolution.

Avoid rebuilding existing solutions; integrate with well-established tools (e.g., Phoenix for tracing).

Do not limit agent interactions to unidirectional communication; enable bi-directional communication.

Do not assume existing microservice discoverability practices are a perfect fit for dynamic agents.

Common Questions

The current AI agent landscape is rapidly growing but also highly fragmented, with many incompatible frameworks. This makes it difficult to discover, switch between, or compose agents from different sources without significant rework.

Topics

Agent Interoperability Communication Protocols BAI Platform ACP Orchestration

Mentioned in this video

organizationAETA group

A group including Glean, LangChain, LlamaIndex, and Cisco that proposed agency and an agent connect protocol.

conceptNatural Language Interaction Protocol

A proposed protocol by academics and researchers, including from IBM, for standards bodies.

softwareBAI

An open-source platform developed by IBM where users can discover, run, and compose agents from any framework.

softwareLangGraph

A framework used to write one of the agents demonstrated, showcasing interoperability with other frameworks.

personKate Blair

Director of Incubation at IBM Research, leading a team focused on disruptive technologies like AI agents.

organizationIBM Research

Research division where Kate Blair works on AI agents and disruptive technologies.

softwareLlamaIndex

A framework mentioned as part of the AETA group proposing agency and agent connect protocols.

companyArise

The company behind Phoenix, an open-source inference solution for telemetry and tracing.

conceptAI agents

The rapidly growing and fragmented landscape of AI systems designed to perform tasks autonomously.

softwareGlean

Found this useful? Build your knowledge library

Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.

Try Summify free