For Asia’s CIOs, 2026 signals the dawn of the Agentic Era—where AI shifts from chatbots to autonomous co-workers. Yet the consumer ease with large language models hides stark enterprise realities: compute demands straining power grids, legacy networks choking on unprecedented traffic, and “vibe-hacking” agents undermining traditional trust models.
By 2027, reasoning models like Claude Mythos will force CIOs to govern agentic swarms, rewriting IAM and latency for a new digital order. This demands that CIOs move from “building AI” to “governing agentic swarms,” fundamentally rewriting the rules of Identity Access Management (IAM) and network latency.
This is the landscape Jeetu Patel, president and chief product officer of Cisco, addressed in a recent PodChats for FutureCIO interview. His message was clear: the transition from building AI to governing “agentic swarms” demands a fundamental rewrite of security, infrastructure, and observability.
From access control to action control
Perhaps the most profound shift Patel articulated is in enterprise security. Traditional models built for human users are ill-equipped for autonomous agents.
“If you take a step back right now, the entire architecture for security is going to have to get completely rethought,” Patel stated. “With a chatbot, if you had a wrong answer, the worst-case scenario is you got some wrong information. With an agent, if they take the wrong action, it could be irreversible.”
Patel drew a sharp analogy: agents are “like teenagers. They’re supremely smart, but they have really no fear of consequences.” This behavioural gap requires a move from access control to action control. Where a human might be granted broad access to email, agents need far more granular, just-in-time permissions.
“What you have to do is action control,” he explained. “Your granularity of permissioning cannot be as coarse-grained as what you do with humans. You have to give access to the agent just-in-time, at just the right level of access, and for just enough time.”
The stakes could not be higher. As Patel put it: “The difference between simple delegation and secure delegation is the difference between bankruptcy of a company and market leadership.”
The data sovereignty imperative in Southeast Asia
For CIOs across Southeast Asia, these security challenges are compounded by a push for digital sovereignty. Countries from Indonesia to Singapore and across the region are tightening data residency rules, demanding that AI infrastructure and models respect local laws and sovereign control.
Patel noted that this “tiers of sovereignty” concept will shape infrastructure decisions. “As countries get more nationalistic, there’s going to be an emphasis on sovereignty of your data centres,” he said. “You have to understand the level of sovereignty that’s required for the use case, and you have to make sure that that sovereignty can be available architecturally by the providers.”
This is not just a compliance issue; it is a strategic one. Control, along with intelligence and cost, forms one of the three key variables when it comes to AI models.
As nations dial up their expectations of control, the ability to deploy sovereign AI infrastructure will become a competitive differentiator.
Preventing digital congestion collapse
The infrastructure to support these agentic systems is equally transformative. Patel cited Cisco studies showing that agents generate five times more network traffic than humans performing the same task.
“Why is that the case?” he asked. “When I talk to you, and I ask you a question, the way that you retrieve your data is not uploading it to the Internet. You’re just retrieving your data from memory. What is an agent? Every single time I pose a query, that skills file and the memory file might actually have to get uploaded to the model.”
With projections of “10 to 100 agents per human” and a potential 2.4 trillion agents working globally, the network demand becomes staggering. “Infrastructure becomes one of the critical prerequisites of success for AI,” Patel warned. “You need to have the right level of power, compute, memory, and network bandwidth. If you don’t have these things, you will not be able actually to have the agents work for you.”
This has led to a re-architecting of data centres themselves. Patel described the evolution from scale-up networking (within a rack) to scale-out (across racks) to scale-across (across data centres). As power constraints mount, data centres are moving to where energy is available, potentially even into suborbital space for 30% more efficient solar power.
“The way you architect those data centres, the kind of chips that you need to have to network them, things like deep buffering—all of that needs to happen in a very coordinated fashion across hundreds of kilometres,” he said.
Observability and the new token economy
With plans to scale agentic systems, old monitoring tools are quickly becoming obsolete. “You can’t protect what you can’t see. You can’t reduce the cost of something that you have no idea what it’s consuming,” Patel said, explaining Cisco’s acquisition of Galileo to build a purpose-built observability stack.
“An organisation’s ability to economically and efficiently use tokens and create tokens in their token factories is going to be a superpower and a strategic differentiator,” he said. “You’ve heard these stories where sometimes someone will get a $25,000 bill that they didn’t intend to get, because the agent just went out and consumed all of this activity.”
Cisco’s observability stack aims to provide visibility into agent behaviour, token spend, and infrastructure utilisation, enabling CIOs to intercept runaway agents and control costs in real time.
A call for AI fluency
As the conversation turned to board-level strategy, Patel offered a stark warning:
“Today people are worried about jobs getting lost to AI. I don’t think that’s the worry you should keep in mind. The worry you should have in mind is there’s a huge difference between someone fluent in AI and someone who is not at all fluent in AI. And that differential is not 10%, it’s not 20%, it’s 10x, it’s 100x.” Jeetu Patel
He urged CIOs to drive a culture of AI fluency across their organisations. “You have to get your people to get comfortable with using it, then get good at it, and then get efficient with it. Don’t try to get efficient when you’re not good at it.”
For Patel, the goal is clear: “Good in a company in the next five years of a company that’s not fluent in AI, it’s going to be very hard to consider the company good.”
The agentic era is not a distant future—it is the reality of 2026, and the choices made now will define the winners and losers of the coming decade.
Click on the PodChats player to hear the direction of Patel’s views on the topics that CISOs and CIOs in Asia are addressing in 2026 and into 2027.
With AI agents “vibe hacking” executive styles, how do we defend enterprises when traditional security fails—and who’s accountable when autonomous agents act without human oversight?
Southeast Asia prioritises data sovereignty. How can CIOs keep AI fast and efficient without illegally moving sensitive data across borders?
AI agents generate 100x more network traffic than humans. With hybrid AI adoption surging, how do we prevent a digital congestion collapse in 2026?
Power grids are strained, yet data centres must act as one. Should CIOs embrace complex distributed clusters or invest in smaller, localised GPU pools?
Agent-to-agent commerce is coming. As consumers and partners start using AI agents to interact with businesses, how can enterprises ensure their backends are “machine-readable” and they don’t become invisible in this new agent-driven economy?
With AI shifting from automation to orchestration, how should ROI be measured—cost savings, or agility and revenue creation?
The challenge of industrialising AI is about moving from isolated pilots to “AI factories”. With plans to scale agentic systems, old monitoring tools are obsolete. Should CIOs in the region be planning to rebuild their telemetry and observability stacks from scratch this year to manage these complex, high-speed interactions?
Our topic is The digital traffic jam of 2026. What are your suggestions for CIOs and the board as they look to embed AI in the way of work?
