Tokenomics defines agentic AI economics

The transition from chatbots to autonomous agents is changing the form of demand itself, and tokenomics — the economics of AI token consumption — is emerging because the defining constraint on enterprise budgets as round the clock inference replaces intermittent usage. Fewer than 1% of potential users are currently deploying agents at scale, leaving enormous headroom for growth in inferencing demand, in response to Jeetu Patel (pictured), president and chief product officer of Cisco Systems Inc.

“There’s lower than 1% of the world that’s using agents,” Patel said. “When you imagine that agents are literally a step function improvement from a chatbot, where you possibly can even have either your personal productivity or your organization’s productivity materially change because of this of agents, I don’t see the way you don’t stay in a continued form of supply shortage for a reasonably very long time period.”

Patel spoke with theCUBE’s Dave Vellante on the AMD Advancing AI event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed distributed inference, tokenomics and the shifting economics of agentic AI deployment. (* Disclosure below.)

Tokenomics emerges because the defining constraint on agentic AI

Unlike chatbots, agents don’t wait for a human prompt — they run constantly and increasingly discuss with other agents, which is pushing a brand new class of machine dedicated purely to agent workloads, Patel explained. Cisco’s answer is Cisco Cloud Control, a unified management plane built to trace that sprawl across every location inference can occur.

“It will possibly exit and have a look at your entire inferencing capability that you could have within the cloud, the entire inferencing capability you would possibly have in a non-public data center, in addition to your endpoint devices, in a single management plane,” Patel said. “Inside that hybrid inference model, AMD handles intelligent routing and compute while Cisco supplies network bandwidth, security and what Patel called tokenomics — visibility into which agents are consuming tokens and whether that usage has grow to be excessive.”

Enterprises are also being pushed toward smaller, task-specific models to maintain costs in check fairly than defaulting to frontier systems for each job. Cisco recently released Antares, a family of open-weight models built specifically to locate vulnerabilities in code without sending proprietary data to the cloud, a direct example of that routing logic in practice, Patel noted.

“That individual announcement doesn’t mean that we actually don’t partner very deeply with Anthropic and with OpenAI,” Patel said. “It just implies that you’re going to want to have [small language models] that you should have control over. If I can find 70% of the vulnerabilities that way, great. And the remaining 30% I can find with a frontier model.”

Ease of use, not compute, stays the most important barrier to broader agent adoption. The technology still has to travel a protracted distance before it reaches extraordinary users the way in which the web once did, Patel noted.

“We’re moving at a reasonably fast pace, but we’re nowhere near the extent of simplicity that’s needed for eight billion people all going out and activating 1000’s of agents,” he said.

Here’s the whole video interview, a part of SiliconANGLE’s and theCUBE’s coverage of the AMD Advancing AI event:

(* Disclosure: TheCUBE is a paid media partner for the AMD Advancing AI event. Neither AMD, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE