As agents become connected, the value of every connected application will rise – provided vendors can work together to let their AI agents work together.
Last week, I attended AWS re:Invent 2024 with 60,000 of my closest friends. We were there to catch up on the latest and greatest in the cloud, particularly AI. One of the more interesting sessions was on the topic of the “Internet of Agents” by Vijoy Pandey, SVP/GM of Outshift by Cisco. For those unfamiliar with Outshift, it’s an internal incubator at Cisco focused on emerging technology in agentic AI, quantum networking, and next-gen infrastructure. As a separate group, Outshift can move at the speed of a startup while retaining access to the resources of a large company like Cisco.
The concept of the “Internet of agents” is a simple one. The AI-based agents found in applications can communicate with each other bidirectionally. Pandey’s definition was “an open, internet scale platform for quantum-safe agent-agent and agent-human communication.”
(More on why the term “quantum-safe” was included is at the end of this piece.)
Agent Sprawl
One might wonder what problem the AI-based agents are trying to solve here; it has not manifested itself yet. Yet I believe generative AI is one of those “game-changing” technologies that will alter almost every aspect of our lives. I predict, over time, that every application we use will have a generative AI interface built into it, like how every app has a search box today. Over time, these agents will go from reactive, where we ask them questions, to proactive, where specific agents push us the contextually important information we need to know.
Consider the implications of this. Today, most workers use several applications – anywhere from half a dozen to over 50. As these apps evolve and add agents, we will face “agent sprawl,” where users will have as many virtual agents as they have apps.
At re:Invent, I attended a session that had the CIO of a major bank participating in it, and he brought up how they’re building virtual assistants for their own apps but also are using Teams Copilot and Salesforce’s agent. Post-session, I asked him what he thinks the future looks like, and he told me he foresees a day when users have a “tapestry” of agents they need to pick and choose from. I followed up my question by asking him what he thinks working in that kind of environment would be like, and he said, “likely chaos.”
Fragmented Knowledge
The numerous agents cause several problems. The first is that the agent or assistant is only as knowledgeable as the data in the application. This can create fragmented insights. As an example, consider the case where a company has a great website that does a best-in-class job of showcasing a poorly built product. The web analytics and sales tools that are used before purchase might show high customer satisfaction scores as they measure pre-purchase satisfaction. Once the customer uses the product, the mood will turn from happy to upset, and the contact center will field calls regarding refunds and repairs. Using the generative AI interface to understand customer sentiment will yield different results.
Also, as the agents shift from reactive to proactive, users will be bombarded with messages from these systems as they look to keep you updated and informed. I expect the apps to have controls, much like they do today, so we can control the interactions with the apps, but most users will keep critical apps on. It would be like a CEO having a team of advisors across every business unit in a company, all whispering in his ear at once.
Interconnecting Agents
This is where the Internet of agents brings value. By interconnecting, these assistants can share information, leading to less but more relevant information. In the scenario outlined above, a product owner or sales leader could be alerted when customer sentiment changes, as these pre-purchase agents communicate with call center agents to provide a holistic picture. This would enable the company to better understand what happened and take corrective action.
Also, this will enable users to work in the applications they prefer but still access information from others. Today, a sales leader can pull data from CRM, contact center tools, sales automation applications, and other systems. The data must be brought together manually and likely correlated by people to find the insights. With the Internet of Agents, AI could perform analytics across multiple systems.
The value can be described using Metcalf’s Law, which states that the value of any network is proportional to the square of the number of connected nodes. A network of two nodes has a value of four, whereas a network with 16 nodes has a value of 256, etc. As agents become connected, the value of every connected application will rise.
To accomplish this, the vendors will need to agree to a set of standards and follow them – this is something Pandey and the team at Outshift are working on. This is where I hope the application providers learn from the sins of the past, as many of them have historically preferred walled gardens.
One example is the UC messaging industry. Slack, Teams, Webex, Zoom, etc., all operate in silos, so a worker can’t send a message in Slack to a Teams user. Imagine how useless text messaging would be if one could only send messages to phones of the same manufacturer. The reality is that when systems are open and standards-based, it creates a rising tide, and everyone wins. A small piece of a big pie is worth far more than most of a small pie.
The Agents Are Coming
One final point: Pandey’s definition did include the term “quantum-safe.” I asked why that was included and was told that if one is building the next generation of secure connectivity, security should be future-proofed. Infusing quantum-safe protocols ensures that quantum nodes are added to the infrastructure and communications are secured even against “store now, decrypt later” attacks. This is consistent with conversations I’ve had with other security companies where their primary concern around quantum computing is bad actors stealing data today, then using quantum to decrypt it at a later date.
To paraphrase Paul Revere, “The agents are coming, the agents are coming,” and I implore the vendor community to get together and standardize the communications between systems and to ensure they are secured. Adoption will be faster, users will be happier, and the value will be greater. Seems like a no-brainer to me.
