Agentic AI Foundation Unveils New Projects and Expansions

While ambassador recruitment typically occurs in June and January, this special intake aims to strengthen local support and engagement in Japan.

This early enrollment reflects the Foundation's commitment to fostering a robust global community for agentic AI development, particularly in key regions like Japan.

This project focuses on secure traffic management between AI agents, Model Context Protocol (MCP) gateways, and large language models (LLMs).

Agent Router was contributed by Tetrate and Bloomberg, highlighting a collaborative effort to enhance the security and efficiency of agentic AI communications.

This move aims to broaden participation and transparency in the development of agentic AI standards.

Members of the public can view the meeting timelines and participate in discussions on critical areas such as authorization, payments, and discovery. This initiative fosters an inclusive environment for community-driven development.

The Foundation's decision to open these groups underscores its commitment to collaborative innovation, ensuring a wider range of perspectives contribute to shaping the future of AI.

This addition reflects the Foundation's growing influence across 14 global sectors.

This new stage expands the Foundation's scope beyond established growth and impact projects.

The shift aims to foster rising stars in the AI domain, providing a supportive environment for innovative ideas in their nascent stages. This move indicates a proactive approach to cultivating future AI breakthroughs.

The 'Sandbox Stage' offers resources and guidance to projects that are still developing, helping them mature and eventually contribute significantly to the agentic AI ecosystem.

David encouraged greater participation from the global community in the Model Context Protocol (MCP) community, especially from the East. He suggested joining the Discord server and participating in various working groups.

This call to action aims to build a default standard for agentic interaction, emphasizing the importance of diverse contributions to achieve a truly global and interoperable AI ecosystem.

A key concern is understanding what transpires within these autonomous systems, particularly regarding agent-to-agent communications.

Questions arise about managing token budgets and monitoring the usage of sub-agents, making effective governance difficult. The current lack of transparency complicates efforts to ensure agents operate within defined parameters.

There is a critical need for security within MCP servers that does not require direct code changes, allowing for dynamic observation and governance without disrupting core functionalities.

Lynn Sun provided a live demonstration of dynamically adjusting token budgets for AI agents at runtime. The demonstration involved increasing a daily token quota from 200, which was deemed too low, to address a failed camera launch.

The policy changes were applied live through the Agent Gateway, showcasing the flexibility and immediate impact of configuration adjustments. This capability allows operators to fine-tune agent performance and resource consumption without system downtime.

This system allows for robust authentication without the need to modify the Model Context Protocol (MCP) server code.

During the demonstration, a JWT token was issued, enabling the publication of a film to a GitHub repository. Authorization checks verified the audience scopes to ensure appropriate access for GitHub API usage.

This approach streamlines security implementation, allowing developers to maintain secure agent interactions efficiently while adhering to established authentication protocols.

Lynn Sun highlighted the ability to configure Model Context Protocol (MCP) authentication policies through the Agent Gateway, eliminating the need for code changes on the MCP server. This policy-driven approach significantly simplifies the management of AI agent security.

Authorization relies on valid tokens and specific audience scopes, ensuring that only authorized agents can perform designated actions. This method enhances security while maintaining operational flexibility.

The demonstration also showcased tracing and log inspection capabilities for tool calls, providing detailed insights into agent activities and aiding in troubleshooting and compliance.

Enterprise systems initially struggled with Large Language Models (LLMs) due to their tendency for hallucinations and reliance on outdated information. This 'grounding gap' posed a significant challenge for accurate and reliable AI applications.

To address this, Retrieval-Augmented Generation (RAG) was introduced to connect LLMs with databases and search engines for up-to-date factual information. Additionally, the industry developed the Model Context Protocol (MCP), which enables function calling for seamless collaboration with external APIs, significantly improving LLM accuracy and relevance.

The proliferation of multi-agent systems led to a fragmented landscape of proprietary, 'home-brewed' protocols, creating an N-times-N sprawl of incompatible communication methods. This lack of standardization hindered seamless interaction between diverse AI agents.

To address this, the Agent-to-Agent (A2A) protocol was initiated to standardize interoperability across different agent ecosystems. A steering committee comprising industry leaders such as AWS, Cisco, IBM, Google, and Microsoft guides its development, ensuring broad industry adoption and robust standards.

The latest production version of the Agent-to-Agent (A2A) protocol, version 1.0, is now available. This version includes signed agent cards that clearly detail each agent's capabilities and the services it can provide.

A2A 1.0 supports multi-protocol binding, accommodating HTTP, JSON-RPC, and gRPC, ensuring broad compatibility across various communication frameworks. It is designed for enterprise-grade deployment, emphasizing reliability and scalability in complex AI environments.

The Agent-to-Agent (A2A) ecosystem is rapidly expanding with the announcement of the AP2 (Agent Payments Protocol) in partnership with PayPal. This new protocol operates on top of A2A, providing secure transaction support for commerce and payment use cases within agentic AI systems.

Over 150 companies currently back the A2A project, demonstrating significant industry support and commitment to standardized agent interoperability. Notable adopters include major players like Huawei and Tencent, further solidifying A2A's global reach.

The growing number of partners and the introduction of specialized protocols like AP2 indicate A2A's potential to become a foundational standard for secure and efficient agent interactions across various sectors.

Attendees are encouraged to enjoy the rest of the conference.