Intelligence must be owned, not rented

Good morning, { AI enthusiasts }. AI is usually described as a “digital teammate,” but the truth is more complex.

Embedding AI systems into core workflows reshapes how teams operate — and raises some tough questions: What actually gets faster? What becomes riskier? And where must humans stay on top of things?

To unpack how this shift is playing out in practice, we sat down with DJ Sampath, SVP of AI Software and Platform at Cisco, on the sidelines of the Cisco AI Summit for a candid take a look at what it takes to construct, secure, and scale with AI within the loop.

In today’s AI rundown:

LATEST DEVELOPMENTS

The Rundown: Cisco sees AI agents as a digital workforce that absorbs routine tasks (like outages), enabling teams to deal with the complex strategic work — and what comes next. The important thing takeaway for corporations? Mastering human-agent collaboration.

Cheung: Cisco has said AI will make the world feel prefer it has “80B people.” What does that mean inside an organization? And the way far can AI go in network ops?

Sampath: For the primary time, we’re deploying digital teammates that may plan, reason, and execute with autonomy. Every leader will manage a constellation of agents working in parallel — investigating, analyzing, remediating — while humans move up the stack to creativity, judgment, and strategic direction.

Inside 12 months, I expect AI to resolve roughly 80% of pattern-based, routine network incidents autonomously. The ultimate 20%, that are multi-vendor, legacy-heavy, or edge-case complexities, will take longer. But identical to self-driving, progress will compound.

Sampath added: Over the following five years, the businesses that learn to design for human–agent collaboration, with trust, governance, and intent on the core, will define the following era of operational performance.

Why it matters: Humans won’t get replaced by AI, but they can be pushed up the stack. As agents absorb the predictable and procedural, the premium will shift to judgment, creativity, and strategic pondering. The winning edge for businesses will come from pairing that human depth with agents’ speed and scale.

The Rundown: Sampath practices what he preaches — using AI to rethink how work gets done every day. From multi-model ideation to coding agents that automate every day briefs, here’s how Cisco’s head of AI actually uses the technology.

Cheung: What form of workflows are you automating with AI? Are you able to give a number of examples?

Sampath: At work, I’ve been experimenting with a straightforward but structured way of using multiple AI tools. First, I separate idea generation from evaluation by drafting a memo, strategy, or customer narrative in a single model, then bringing it right into a second model to critique and improve it. This helps me think more clearly and produce stronger work.

Next, I exploit Cursor to store context in markdown files and folders that AI can reference. Over time, this builds a knowledge base, like a long-term thought partner that understands my frameworks and past work.

Sampath added: I also connect AI to my calendar and meeting notes to review context before customer, partner, or analyst conversations. Plus, I’ve began using coding agents to automate work like every day briefs, product reviews, and document evaluation.

Why it matters: Sampath’s approach shows that the long run of labor can be defined by how teams stitch together agents, models, and systems into structured but adaptable workflows. The Cursor-as-knowledge-base idea is particularly actionable, turning one-off AI interactions right into a compounding system that gets smarter over time.

The Rundown: Most enterprises are held back from AI adoption not by a scarcity of ambition, but by infrastructure debt and siloed data. Sampath says the true unlock requires pairing modern infrastructure with leadership clarity — and embedding intelligence directly into products.

Cheung: Cisco’s AI Readiness Index shows only 28% of organizations imagine they’re ready for AI workloads. What’s holding back the remaining, and what does it take to be a real AI company today?

Sampath: What’s holding back the opposite 72% isn’t just missing GPUs. It’s AI infrastructure debt: legacy networks, fragmented data, siloed tooling. Systems built for yesterday’s applications can’t support the throughput, real-time processing, and autonomy that modern AI demands.

One other key component is the necessity to pair modern infrastructure with leadership clarity — governance, strategy, and alignment to business outcomes. The leaders have to unravel for each, defining tech and the way work will occur with AI.

Sampath added: The sustainable advantage will come when intelligence is embedded into the product itself. When the model is trained in your contextual enterprise data, it improves constantly and directly drives outcomes. So, the product becomes the model, and the model becomes the product.

Why it matters: Being “AI-ready” means rethinking the entire stack, from infra to security to the appliance layer. Corporations that make AI core to their product (not only a feature) can unlock feedback loops backed by proprietary data, where outcomes improve constantly and enable them to maneuver faster.

The Rundown: Sampath says that probably the most urgent AI security threat is the chance of agent compromise. As these systems tackle more tasks, they develop into powerful attack surfaces — with the chance only growing with time.

Cheung: What’s probably the most concrete, real AI security threat immediately?

Sampath: Probably the most immediate AI security risk is the compromise and misuse of autonomous agents. As enterprises deploy agentic systems that access data, invoke tools, and make decisions independently, those agents develop into a brand new attack surface. They could be hijacked, impersonated, or manipulated to exfiltrate data or execute unauthorized commands at machine speed.

We’re already seeing attackers probe these gaps. That’s why we consider AI security from two perspectives: protecting the enterprise from agents and protecting agents from the world — with zero-trust identity, control over agent protocols and gear registries, and continuous behavioral monitoring.

Cheung: If an enterprise is moving from AI pilots to production, what’s the primary system they should harden? And where should a human at all times stay within the loop?

Sampath: The very first thing to harden is the agent infrastructure. The true risk sits within the connective tissue: the protocols that link agents to tools, data, and one another. Standards like Model Context Protocol and agent-to-agent have develop into the backbone of autonomous workflows, but they scaled faster than the safety around them.

As teams roll out agentic AI, anything that affects trust, access, or control over critical systems — granting privileges, changing production environments, authorizing sensitive data access, initiating irreversible actions — should never run fully autonomous. When consequences are real, accountability must be human.

The fitting model isn’t human-out-of-the-loop. It’s AI-in-the-loop. Let agents handle the routine and low-risk at speed, and keep humans because the authority where impact is high.

Why it matters: The shift from models that answer to agents that act introduces a brand new class of risk — systemic, fast-moving, and hard to contain. To secure operations on this future, organizations may have to bake their agentic deployments with identity, guardrails, and constant oversight, treating them like real entities.

The Rundown: Many corporations are racing to bolt AI onto existing products, but Sampath argues the true moat comes from embedding intelligence into the product itself — and makes a daring case that the long run of AI shouldn’t be controlled by a handful of centralized providers.

Cheung: You’ve said “corporations which might be a skinny shim on top of a model — their days are numbered.” What does that really mean?

Sampath: We mean this: adding a generative API to an existing product isn’t a technique — it’s a feature.

Sustainable advantage comes when intelligence is embedded into the product itself. When the model is trained in your contextual enterprise data, it improves constantly and directly drives outcomes. That closed loop — fueled by proprietary machine data — is the moat.

Cheung: If you happen to weren’t at Cisco, what AI problem would you ought to work on?

Sampath: The issue I care most about is ownership of intelligence. I don’t imagine the long run belongs to a handful of centralized models controlled by a number of providers. I imagine intelligence must be owned by enterprises — and ultimately, by individuals.

Which means constructing a full stack that permits organizations to develop, fine-tune, deploy, and govern models on their very own terms.

Why it matters: “Intelligence must be owned, not rented” is an interesting thesis when most enterprises currently leverage centralized AI providers. It raises an issue every company must be asking: are you constructing AI capabilities that compound over time, or renting them from someone who could eventually change the terms?

GO DEEPER

Watch all the sessions from the Cisco AI Summit on-demand, featuring discussions including:

Click here to see all of the sessions and watch on-demand.

See you soon,