Vinton Cerf, the man widely called a “father of the internet,” is stepping down from his role as Google’s chief internet evangelist next week, according to TechCrunch AI. The news broke at the Open Frontier conference hosted by the Laude Institute, where Cerf appeared by video feed and got a round of applause from the room. He’s 83, he’s been at Google more than 20 years, and he leaves behind a career that shaped the network we all run on.
But the retirement isn’t the real story here. What stands out is the prediction Cerf made on his way out, and why it lands at exactly the right moment for anyone building AI products.
The prediction that matters
Cerf, who co-designed TCP/IP with Robert Kahn in the 1970s, thinks the next wave of AI is about to repeat the early internet’s playbook. As he told the panel, per TechCrunch AI: “The agentic model of AI, with multiple agents from multiple sources interacting with each other, is going to force composability, and a requirement for interoperability and standardization.”
Translate that out of engineer-speak: once you have thousands of AI agents from different companies trying to work together, they need a shared rulebook. The same way different computer networks needed TCP/IP to talk to each other, agents will need agreed standards to coordinate without falling apart.
And Cerf doesn’t think plain English will cut it. Other panelists figured agents could just chat in natural language. He pushed back hard:
- “I don’t think English is going to be the best choice. There’s a flexibility in it, but there’s ambiguity.”
- Precision matters because “an agent really needs to be sure the other agent understands what it is that they just agreed to do together.”
- His warning image: the telephone game, where a whispered message turns to nonsense after ten people. “Imagine a bunch of agents talking to each other in natural language, you know, that’s kind of terrifying.”
When the guy who designed the internet’s core protocols tells you natural language won’t scale for machine-to-machine coordination, that’s worth writing down.
Why this matters now
The conference kept circling one tension, TechCrunch AI reports: advanced AI models are bunching up inside a few well-funded labs, while the open internet that made Cerf’s protocols durable was built on decentralization. That contrast is the whole game for the next few years.
Here’s the strategic angle. If Cerf is right, whoever defines the agent interoperability standards early gets outsized power over how the entire agentic economy runs. That’s not theory. It’s precisely what happened in the early internet protocol wars, and the winners set the terms for decades.
We’re already seeing the opening moves. Anthropic’s Model Context Protocol, Google’s own agent-to-agent efforts, and a scramble of open specs are all land grabs for the same territory Cerf is describing. The standard that wins becomes the rails everyone else is forced to ride.
What to do about it
If you’re building or betting on AI right now, a few practical takeaways:
- Watch the standards, not just the models. The model you use will change every few months. The protocol your agents speak could lock in for years. Track MCP and the emerging agent-communication specs like they’re infrastructure decisions, because they are.
- Don’t build your agent stack on English glue. If your multi-agent system passes instructions around in loose natural language, Cerf’s telephone-game problem is coming for you. Favor structured, typed, verifiable message formats where precision counts.
- Bet on composability early. Founders leaning on open infrastructure for the next wave of products are positioned for exactly the interoperability shift Cerf sees coming. Closed, one-off integrations will age badly.
Cerf’s track record on this stuff is about as good as it gets in computing, complete with a Turing Award and a Presidential Medal of Freedom. So when he calls the shape of the next decade on his way out the door, the smart move is to plan around it.
The man who helped make networks talk to each other is retiring just as the industry faces the same problem all over again, this time with agents. More detail on his remarks and the Open Frontier panel is available at the original TechCrunch AI report.