For the last two years, most conversations about artificial intelligence have focused on models. Which model is leading, which vendor has the most advanced capabilities, and which platform organizations should standardize on.

Those are important discussions, but they aren’t the conversations we’re having with customers.

What we’re seeing is a growing realization that AI adoption is becoming an infrastructure challenge. As organizations move beyond simple chat interfaces and begin deploying autonomous agents capable of executing workflows, accessing systems, and making decisions, entirely new requirements emerge around networking, security, governance, and operational visibility.

On Tuesday I watched a Bloomberg  interview with Cisco President and Chief Product Officer Jeetu Patel, which reinforced this view. His central argument was not really about AI models at all. It was about what AI does to enterprise infrastructure.  

The Shift From Chatbots To Agents

Most organizations are familiar with AI as an assistant. Employees ask questions, generate content, summarize information, or analyze data.

The next phase is different.

According to Patel, we’re moving from human-driven interactions to machine-driven workflows, where autonomous agents execute tasks with minimal human intervention.  

That shift has significant implications for enterprise IT.

When an employee accesses a system, there are established controls, monitoring capabilities, and governance processes. When thousands of AI agents begin interacting with applications, databases, APIs, and other agents, the scale and complexity increase dramatically.

The challenge is no longer simply deploying AI.

The challenge becomes trusting it.

This is why security, governance, and observability are quickly becoming the most important conversations surrounding enterprise AI adoption.

Security Will Determine The Pace Of AI Adoption

One of the strongest themes throughout Patel’s interview was the importance of trust. Organizations are increasingly willing to delegate work to AI systems, but only if they can understand what those systems are doing and ensure they are operating securely.  

At Hypershift, we’re seeing the same concern from customers.

Very few organizations question whether AI can create value. Most already know it can.

Instead, they want answers to practical questions:

• How do we secure AI agents?

• How do we control access to sensitive systems?

• How do we monitor behavior?

• How do we prevent unintended actions?

• How do we maintain compliance and governance?

These are not model-selection problems.

They are infrastructure problems.

Cisco’s strategy appears to recognize that reality. Rather than competing directly in the race to build foundation models, Cisco is positioning itself as the layer that secures, governs, and connects the AI ecosystem.

Why Project Glasswing Matters

One of the more interesting topics discussed during the interview was Cisco’s participation in Anthropic’s Project Glasswing.

The initiative provided selected organizations with access to advanced AI capabilities before broader public availability, allowing them to identify vulnerabilities, exposure points, and potential infrastructure risks. Cisco used that opportunity to discover weaknesses, develop mitigations, and strengthen protections before those capabilities became widely available.  

The significance of this approach should not be overlooked.

Historically, security teams often learn about new attack techniques after they begin appearing in the wild. Programs like Project Glasswing create an opportunity to understand potential risks before adversaries gain access to the same technologies.

For organizations operating critical infrastructure, healthcare systems, financial platforms, or other high-value environments, that head start can be extremely valuable.

The Most Practical Innovation: Live Protect

While much of the industry focuses on future AI capabilities, one of the most practical announcements discussed by Patel was Cisco’s Live Protect technology.

Every IT leader understands the challenge of vulnerability management. A new vulnerability is disclosed, a patch becomes available, and the organization begins testing before deployment. Depending on the environment, that process can take days or even weeks.

Unfortunately, attackers don’t wait.

Live Protect is designed to provide immediate protection during that gap, creating a virtual shield while organizations work through their normal remediation process. Patel described protection being available within minutes of disclosure rather than waiting for a traditional patching cycle to complete.  

For many enterprises, that type of operational security capability may prove more valuable than any single AI feature announcement.

The Hidden Challenge: AI Economics

Another topic receiving far less attention than it deserves is AI cost management.

As organizations deploy agents and automate more workflows, token consumption becomes a measurable operational expense. Without proper governance, costs can increase rapidly.

Patel discussed Cisco’s work around token monitoring, utilization tracking, agent behavior analysis, and infrastructure efficiency. The goal is not simply to use AI, but to understand how AI resources are being consumed and whether those resources are generating value.  

This feels remarkably similar to the early days of cloud computing. Organizations initially focused on adoption. Later, they realized they needed FinOps practices to control costs and optimize usage.

AI appears to be following the same path.

The enterprises that succeed will be the ones that treat AI as an operational discipline rather than an experimental project.

The Networking Supercycle Is Already Beginning

Perhaps the most important statement from the interview was Cisco’s research suggesting that agentic workloads generate approximately 450% more network traffic than equivalent human-driven activities.  

If that estimate proves directionally correct, the implications are substantial.

Unlike employees, agents operate continuously. They communicate with applications, APIs, databases, cloud platforms, and other agents around the clock. Every interaction generates network traffic, and every workflow creates additional infrastructure demand.

This is why Cisco describes AI as the beginning of a networking supercycle rather than a threat to networking.

More AI means more connectivity requirements, more visibility requirements, and more infrastructure to manage.

The Future Is Distributed

The final prediction from Patel may ultimately be the most important.

Today, most AI inference occurs in centralized cloud environments. Over time, however, Cisco expects inference to become increasingly distributed across laptops, workstations, AI-enabled PCs, and localized computeresources. Patel referred to this trend as “desk-side computing.”  

The economics make sense. As local hardware becomes more capable and organizations seek to reduce token and inference costs, more workloads will move closer to users and data sources.

That creates new demands for security, networking, device management, observability, and governance.

In other words, the very areas where Cisco and Hypershift already help customers operate today.

What This Means For Enterprise IT Leaders

The biggest takeaway from Cisco’s AI strategy is that AI should not be viewed as a standalone technology initiative.

Every successful AI deployment eventually becomes a networking project, a security project, a governance project, and an infrastructure project.

The organizations that realize this early will be better positioned to scale AI safely and effectively.

At Hypershift, that’s where we’re focused. Through our partnership with Cisco, we’re helping customers build the secure, resilient, and observable infrastructure required for the next generation of AI-enabled operations.

The future of AI will not be determined solely by models.

It will be determined by the infrastructure that makes those models usable, secure, and scalable.