Partner Content

Panel Recap: Building the Internet of Agents

Panel Recap: Building the Internet of Agents
By
The Information Partnerships

Since ChatGPT’s 2022 debut, businesses have envisioned generative AI creating not just chatbots, but also automated tools capable of complex, multi-step tasks like managing retail returns, generating research reports, and planning travel.

This spring, new open-source protocols emerged to facilitate AI agent collaboration. Google introduced the A2A Protocol, while the AGNTCY collective—founded by Cisco, LangChain, and Galileo—launched an infrastructure framework for agent-to-agent collaboration. During a virtual panel discussion, The Information reporter Kevin McLaughlin discussed how these emerging standards are reshaping the AI landscape with three industry leaders:

  • Vijoy Pandey, Ph.D., general manager and senior vice president, Outshift by Cisco (core maintainer for AGNTCY)
  • Rao Surapaneni, vice president and general manager of business application platform, Google Cloud
  • Jesse Zhang, co-founder and CEO, Decagon

From Communication to Collaboration

Pandey believes the ultimate goal is for AI agents to function like human teams. “They’ll be really good at certain tasks, and they’ll all need to come together, get discovered, communicate, collaborate and solve a business need,” he said. “They’ll all come from different vendors, sit on different clouds and take on different personas.”

While standards like Anthropic’s Model Context Protocol (MCP), introduced in late 2024, primarily enable communication between agents and data sources, Pandey emphasized that the “internet of agents” will demand more. Agents will need to discover each other, manage authorization and access, and even evaluate capabilities. Google Cloud’s Agent2Agent and Cisco’s AGNTCY are designed to enable this deeper, more interactive connectivity.

Surapaneni elaborated, “If you are a single developer or a single department that has full access to the data and the tools, then MCP works great. But when we are crossing these chasms or these silos, that’s where we’re incorporating A2A.”

Current Capabilities

Surapaneni described coordinated agentic AI as “in flight.” Google has published several real-world use cases for its A2A protocol, including travel planning and candidate sourcing, where AI agents collaborate on tasks like forwarding suggestions to hiring managers and conducting background checks. “There are companies that are implementing it, consumers seeing it and internal employees using it,” he said.

Pandey noted that Cisco engineers developed and now use an agent called SRE Jarvis, now part of the CNOE (Cloud Native Operational Excellence) initiative. This agent can access multiple large language models and cloud applications, assisting engineers with platform engineering tasks such as CI/CD and resource creation in AWS and GCP. “We’ve gone from 49 days to 16 minutes on certain tasks,” Pandey stated, adding that they’ve automated 30% of their tasks. They’ve also created agents that automate validation and testability for network configurations.

Zhang acknowledged the “exciting” prospect of multi-agent workflows for tasks like reordering a lost credit card or disputing a charge but noted that most such projects are still in the pilot phase. His company, Decagon, offers an AI agent for customer support, though most inter-agent coordination currently relies on traditional application programming interfaces (APIs). However, he sees the benefit of newer protocols: “As people build more agents, you could benefit a lot from using one of these [newer] protocols. You save a lot of steps, and you’re able to cover much more complexity.”

Looking Ahead

Zhang observed that while AI agent demos are “fairly impressive,” real-world implementations often underwhelm or fail. A key issue is that the internet is predominantly browser-based, designed for humans. Agentic and multi-agent workflows will be more successful, Zhang believes, when agents can directly interact with underlying data rather than navigating websites. “You don’t really want the agent to be pulling up your DoorDash app and doing things in the [user interface],” he explained.

Ultimately, Pandey stressed the importance of vendor collaboration to foster interoperability, anticipating partnerships between companies like Cisco and Google to realize the internet of agents. “Complexity and heterogeneity help no one—not the vendors, not the operators, not the developers,” he asserted. “The faster we converge on an architecture, the better it is for everyone.”

Editor’s Note: After this event was recorded, Google announced the move of their A2A Project to the Linux Foundation, with Cisco as a founding member of the new steering committee alongside AWS, Microsoft, Salesforce, SAP, and ServiceNow.