The New Infrastructure for AI Agents: Why Config Automation Alone Isn’t the Answer

In a recent article from ZDNet, Cisco unveiled its Deep Network Model, a large language model (LLM) designed to power AI agents and automate network operations. While this marks a significant shift toward intent-based automation, the bigger question remains: Can traditional networking models—even when automated—support the rapidly evolving world of AI agents, inference, and real-time B2B data interactions?

Agentic AI Microservices: A Paradigm Shift

As someone experienced in MPLS, IP backbone design, and SD-WAN, I see the Wide Area networking industry is heading in the same direction hyperscale cloud providers did years ago: transitioning from monolithic, static architectures to microservices and ephemeral agents.

In this new model

• Billions of AI agents will be spun up across distributed cloud regions.
• These agents use private IPs and need secure access to enterprise data or APIs—often sitting behind firewalls or private networks.
• Connectivity must be real-time, secure, and scale dynamically, without exposing endpoints to the public Internet.

The question is not just about “automating” configuration—it’s about reimagining the network

What Cisco’s AI Configuration Agent Brings

Cisco’s new AI agent can automate configuration tasks like tunnel creation, BGP setup, and route monitoring. In theory, this allows rapid onboarding of sites or AI endpoints. But the reality is:

“Automation applied to an inefficient operation will magnify the inefficiency.”
— Bill Gates, “The Road Ahead” (1996)

Merely automating legacy constructs like MPLS, IPsec and SD-WAN tunnels doesn’t solve the fundamental problem: scale and dynamism.

AI agents don’t live in fixed branches—they spin up across clouds, edge nodes, and inference clusters. Their network needs are:

• Ephemeral
• Dynamic
• Policy-Intensve
• Latency and QoS-Sensitive

Why MPLS and SD-WAN Fall Short Even with Automation

• MPLS Cannot handle short-lived agent IPs.
• SD-WAN/IPsec meshes don't scale gracefully—each connection adds exponential routing and policy overhead.

 What’s Really Needed? A New Private Internet Fabric

This isn’t just about automating traditional networks—it’s about evolving our routing infrastructure designed for:

• Private AI Agent-to-Enterprise connectivity
• Dynamic route policy – Where agents self-discover, self-register, and connect without static tunnel provisioning.
• Data assurance – Visibility not just in the data center, but across every path and policy flow.

Conclusion: Time to Build the Right Infrastructure, Not Just a Faster Horse

While Cisco’s automation tools provide incremental improvements, they ultimately reinforce legacy networking paradigms—paradigms that no longer meet the demands of modern enterprises.

In an era of agentic AI, ephemeral workloads, and distributed intelligence, the objective is not to optimize yesterday’s architecture—but to design tomorrow’s.

For agent-based AI to truly deliver value, the network must evolve.

It must become:

• Sovereign – with full control over data location, flow, and compliance
• Self-service – instantly programmable and intent driven
• Self-adapting – intelligent enough to respond in real time without human intervention

That’s why we deliver data infrastructure services—engineered for a world where data is everywhere, sovereignty is non-negotiable, and AI is the next operating layer of business.

The foundational technologies are already here—BGP, MPLS, Segment Routing—but they need to be repurposed with intent-based design, cloud-native principles, and zero-trust security at the core.

The future lies in connecting AI-native services through publish-subscribe topologies with centralized enforcement points. This model empowers enterprises to maintain control, enforce compliance, and ensure data sovereignty—while unlocking the full potential of distributed AI.  Let’s not settle for automating yesterday’s complexity.