In the rapidly evolving landscape of 2026, the shift from simple conversational bots to fully autonomous agentic AI has fundamentally redefined the enterprise technology stack. Dominic Jainy, a veteran in artificial intelligence and cloud infrastructure, joins us to discuss how the integration of Cisco and Splunk is tackling the “crisis of trust” currently facing these digital teammates. This conversation explores the transition of inference as the dominant workload, the move toward edge-based hybrid models, and the critical necessity of deep observability in a world where agents now significantly out-consume human users in token volume. We delve into the complexities of managing “teenager-like” AI agents and the architectural shifts required to maintain security in a post-Mythos world.
With the transition from simple chatbots to autonomous agents acting as teammates, what specific infrastructure changes are required to handle these new and constant workloads?
The shift we are seeing is essentially the move from spiky, human-gated interactions to a world where inference is the primary workload. In the past, even the most dedicated power users couldn’t break the infrastructure because their engagement was limited by their own capacity to type and think, but agents operate twenty-four-seven and are constantly spawning new workloads for other agents. This year has been a massive turning point; in February, the tokens consumed by agents exceeded those consumed by humans for the first time in history. Just seven months later, agents are consuming five times as many tokens, creating a demand pattern that is both higher and far more consistent than anything we’ve seen before. To support this, infrastructure must move away from being a passive resource to a dynamic environment that can handle continuous, high-volume throughput without hitting bottlenecks that would stall autonomous processes.
You’ve noted that agents are often described as having the intelligence of a teenager but with poor judgment; how does this perspective change the way we approach enterprise observability?
When we think of agents as teenagers, we recognize that they are incredibly smart and fearless, yet they lack the seasoned judgment to avoid risky behaviors or subtle “drift” in their logic. Because 60% of global AI compute capacity is now dedicated specifically to inference, we are dealing with billions or even trillions of agents that utilize over five times more network bandwidth than human employees. This scale makes it impossible to manage them through traditional manual oversight, so visibility becomes our only real lever for control. You simply cannot fix what you cannot see, and in an agentic environment, you need real-time telemetry to catch errors before they cascade through the network. We are moving toward a model where observability isn’t just about checking if a server is up, but about monitoring the behavior and decision-making patterns of these digital teammates to ensure they remain within safe operational boundaries.
As tokens become the primary currency of the digital economy, how are organizations restructuring their data centers and network edges to manage these ballooning costs?
The cost of tokens is escalating so rapidly that organizations can no longer rely solely on hyperscale data centers to carry the load; it is becoming a matter of financial survival to find more efficient ways to process information. We are seeing a significant trend toward open-source and open-weight AI models that allow for more localized control and lower overhead. This is driving a hybrid infrastructure model where inferencing happens at the network edge, in on-prem data centers, and even on high-powered desktop computers, rather than sending every request back to a central cloud. By processing data closer to where the agent is actually working, companies can reduce the massive bandwidth strain and keep token costs from spiraling out of control. It’s a strategic shift that prioritizes decentralization, allowing the network to act as a cohesive, distributed brain rather than a series of pipes leading to a single source.
In this “post-Mythos” era of AI, how does the vertical integration of silicon, network, and security layers change the way we respond to cyber breaches or system outages?
In our current reality, every single action an agent takes is simultaneously a routing challenge, a trust decision, and a telemetry event, which means security can no longer be an afterthought or a separate layer. By vertically integrating the entire stack—from the actual silicon and photonics up through the network, data, and the agents themselves—we create an environment where detection and investigation become almost instantaneous. When the worst-case scenario occurs, such as a technical outage or a sophisticated cyber breach, a unified platform like Cisco Cloud Control allows us to correlate telemetry from multiple domains simultaneously. This level of federation means we aren’t hunting through siloed data; instead, we have a centralized control plane that provides the digital resilience needed to remediate issues before they impact the broader business. It’s about building a moat of control where safety and security are baked into the very fabric of the infrastructure, rather than being applied as a patch.
What is your forecast for the evolution of AgenticOps over the next few years?
I believe we are heading toward a future where the distinction between “IT operations” and “AI management” will vanish entirely, replaced by a unified discipline of AgenticOps. We will see the emergence of highly specialized agentic security operations centers (SOCs) that possess the native ability to process and understand every single security event in real-time, far beyond the speed of any human analyst. Organizations will stop viewing AI as a tool and start treating it as a dynamic workforce that requires its own set of labor laws, performance reviews, and safety protocols, all enforced through automated telemetry. Ultimately, the winners in this space will be those who can harness the massive surge in token consumption without losing visibility, turning the “teenager-like” unpredictability of current agents into a mature, resilient, and fully transparent digital workforce. This will lead to a standard where every enterprise action is backed by an ironclad audit trail, ensuring that even as agents become more autonomous, they remain fundamentally aligned with human intent and corporate safety.