Steve Saunders:
For the past two years, the technology industry has been obsessed with AI agents, digital workers, autonomous systems, and software capable of making decisions independently. But the most important question may not be whether agents can think. It may be what happens when millions of them begin interacting with one another.
Steve Saunders:
Because once autonomous systems start making decisions at scale, intelligence stops being the primary challenge. Coordination becomes the challenge and coordination has always been a telecom responsibility. Welcome to Carrier 2.0.
Masum Mir:
Super chatty, always chatty. We think it is this chattiness, multiple chat all colliding together. It's going to create a very interesting traffic pattern.
Steve Saunders:
The first phase of AI generated content. The second phase generates action. Agentic systems do not simply answer questions. They make decisions, initiate workflows and increasingly interact with other agents.
Shazia Sobani:
They are chatty, but if you can use that chat to your benefit and agentic AI actually allows us to do that where it can pick up the sentiment, it can pick up the trends, it can create insights out of that.
Kelly Ahuja:
For each human, like when you're doing some nice AI project, you actually may be spawning 80 to 100 non-human identities according to Gartner. Agents aren't just confined to cloud. They're going to be everywhere and they're going to be talking to everything else. They're going to be accessing data everywhere else.
Steve Saunders:
The implications are profound. If every employee application and workflow can spawn dozens of autonomous identities, governance quickly becomes more important than intelligence.
The industry is focused on building agents, but the harder problem may be governing the interactions between them.
Steve Saunders:
Every autonomous system is trying to optimize an outcome. Performance, security, efficiency and business objectives don't necessarily align. As autonomous systems begin pursuing competing goals simultaneously over the same network, new forms of complexity will emerge.
Kelly Ahuja:
You have to define the business policy around that non-human identity, how it's associated with you as a human, if you're going to be in the loop or are they autonomous.
Steve Saunders:
Traditional software follows instructions, agentic systems, negotiate outcomes. That distinction changes the architecture entirely. Imagine thousands of autonomous systems operating simultaneously across a network. Each individual action may be rational. The challenge emerges when those systems begin responding to one another or talking over each other. The risk is not necessarily a rogue AI. The risk is emergent behavior created by large numbers of autonomous systems interacting at machine speed.
Masum Mir:
The moment we start to see more and more agents and this whole idea of agents and agents of the agents and the super agents, someone has to understand the full environment. It's like the supervisor that you will have in a production factory. So who is that supervisor? This is the super agent that needs to connect all these sensory data from one place and make them work in harmony.
Steve Saunders:
At that point, visibility becomes critical. Someone, or something, must understand the broader environment in which those decisions are being made. So this is where telecom becomes unexpectedly relevant. Telecom has spent decades coordinating distributed systems through routing, policy, orchestration and control plane technologies. Long before the industry started talking about AI agents, telecom was solving coordination problems at enormous scale.
Gurudatt Shenoy:
The network has to be prepared, and this is what we are building for. Well, these are new characteristics on the network driven by AI that we haven't seen to this scale before and it's coming really fast at us.
John Keib:
We're over-provisioning our network pretty significantly to prepare for that. We're starting to see usage in Q3 and Q4 pop to really unprecedented levels.
Steve Saunders:
Telecom's next challenge may not be moving traffic. It may be governing machine speed decision-making.
Steve Saunders:
This is my new hobbyhorse. I call it operational sovereignty or OPSOV. Historically, sovereignty referred to political authority. In the AI economy, however, sovereignty increasingly becomes operational. And the question is no longer simply who owns the infrastructure. The question is who controls the decisions made across that infrastructure?
Logan Wolfe:
It's not something that one can buy as a SKU. You need to design for sovereignty as a strategy for business as optionality. And then the technology is a tool to get that done.
Kelly Ahuja:
Sovereignty is not about a geographic place. Sovereignty is about control and where that control resides.
Steve Saunders:
As intelligence becomes distributed, sovereignty becomes much less about geography and much more about governance.
For most of its history, the network transported information. Increasingly it coordinates systems.
Andrew Feldman:
In an industrial AI economy, it's primarily machines reading the data, primarily machines acting on the answers.
Kelly Ahuja:
The question, though, is what's the best approach of implementing these, what you call governance or the controls and where do you do it? The enforcement layer. Where do you put your enforcement layer? You want to know what's going on in your environment and that environment is your network. You got to see it in line. You got to say who's talking to what and what's accessing what data.
Steve Saunders:
And this may become the defining architectural question of the AI economy, not where intelligence resides, but where governance resides and not where the agents operate, but where the enforcement layer operates.
Blair Levin:
We're here so that 10 years from now we can say, but for what we did, there would be a pothole on the information highway, but we fixed it before it and it was apparent to anybody. And that's what we should be doing on AI now.
Steve Saunders:
Our industry often describes agentic AI as the next phase of automation and that's true, but the biggest story may be control. As autonomous systems begin making decisions across networks, applications and infrastructure, the central challenge shifts from intelligence to governance. The question is no longer whether machines can make decisions. It is how those decisions are coordinated. Who establishes the policies that guide them and where the authority to govern them ultimately resides? In other words, who governs the agents? Now that question leads directly to orchestration, policy and control, and it leads to the control plane of the AI economy. And that's why operational sovereignty may become one of the defining concepts of the next decade.
Hi there. If someone shared this episode with you, that means they like you. Please hit subscribe and catch up on all the episodes.