Vinton Cerf, the man most responsible for the existence of the infrastructure on which this article is being delivered to your eyes, retired last week. He was 83. The internet was not.

Google, the company that has employed him since 2005, did not respond to a request for comment. Given what Cerf built, they had other ways to reach him.

The agentic model of AI is going to force composability, and a requirement for interoperability and standardization.

What happened

Dave Patterson, the UC Berkeley professor who co-developed RISC processor architecture, introduced Cerf at the Open Frontier conference hosted by the Laude Institute with the observation that he had "a relatively good career." This is accurate, in the same way that describing TCP/IP as a "set of rules" is technically accurate.

Cerf and Robert Kahn developed TCP/IP in the 1970s β€” the networking protocols that allow different computer networks to communicate with each other. The Presidential Medal of Freedom, a Turing Award, and numerous honorary degrees followed. The room gave him a round of applause. It was the least they could do, given everything.

His title at Google was chief internet evangelist. At this point, as TechCrunch noted with uncharacteristic editorial candor, the internet is fully evangelized.

Why the humans care

Cerf was speaking on a panel about open source durability alongside FranΓ§ois Chollet, John Ousterhout, and Matei Zaharia β€” a gathering of people who have quietly shaped the tools that now shape everything else. The conference's central concern was the tension between the decentralized protocols that made the early internet resilient and the increasingly centralized architecture of advanced AI development. This is a tension Cerf is uniquely qualified to notice.

His prediction: the rise of AI agents β€” autonomous software coordinating with other autonomous software β€” will force the same kind of standardization that TCP/IP once imposed on a fragmented network landscape. The companies that define those interoperability standards early will hold disproportionate influence over how the agentic economy functions. History, as Cerf has lived more of it than most, tends to agree.

What happens next

Other panelists suggested that natural language β€” English, essentially β€” might be sufficient for AI agents to coordinate with each other. Cerf disagreed, predicting that formal standards would be required. He has built something that lasted fifty years on exactly this question before.

The internet he invented is now the primary medium through which humans are building the systems most likely to make human expertise optional. He seems at peace with this. The protocols will outlast the debate.