The man who wrote the rules that let a billion networks talk to each other says AI is about to repeat the internet's mistakes, unless it learns from them first.
The Summary
- Vinton Cerf, co-inventor of TCP/IP protocols, compared AI's current trajectory to the early internet at the Open Frontiers conference, warning that three principles will determine whether AI reaches its potential.
- Cerf is stepping down as Google's chief internet evangelist at 83, timing his departure as the technology he helped create faces its biggest inflection point since Web2.
- His core lesson: the internet only became ubiquitous because no single company owned it, and that openness is now AI's fork-in-the-road moment.
The Signal
Cerf's timing matters. He is leaving Google just as the company and its peers race to build proprietary AI systems, closed models, and walled agent ecosystems. His parting message is not subtle.
At the Open Frontiers conference, Cerf argued that open standards made the internet work. "In the case of internet, it only worked because it was going to be distributed to begin with. And so we left the rules very open. We just said if you can find somebody to connect to and you follow the rules of the protocols, it should work."
That openness let a university network in California, a government research lab, and a commercial ISP all connect seamlessly. No permission required. No licensing fees. No platform lock-in.
"We left the rules very open. If you can find somebody to connect to and you follow the rules of the protocols, it should work."
Now compare that to AI in 2026. OpenAI's models talk to OpenAI's tools. Anthropic's agents work best in Anthropic's environment. Google's AI lives in Google's cloud. The pattern is clear: every major AI lab is building a moat, not a protocol.
Cerf's three principles from the internet era map directly to AI's current crossroads:
- Open standards beat closed systems
- Distributed architecture scales better than centralized control
- Interoperability creates more value than exclusivity
The irony is thick. Cerf spent two decades at Google evangelizing the open internet while the company built some of the most powerful closed systems in history. Search, Maps, YouTube: all run on open protocols but controlled by a single entity.
Now AI is repeating that pattern at 10x speed. The models are closed. The training data is proprietary. The agent frameworks do not talk to each other. We are building Web2 again, but with more compute and higher stakes.
The Implication
The Fourth Web depends on AI agents that can work together across platforms, companies, and ecosystems. If every AI lab builds its own walled garden, we get Web2 with chatbots. If they follow Cerf's playbook and standardize protocols for agent communication, memory sharing, and task handoff, we get something new.
Watch for the first serious attempt at an open agent protocol. When it happens, it will matter more than any single model release. The internet won when TCP/IP beat out proprietary alternatives. AI will face the same choice. Cerf just told us how it ends if we pick wrong.