Ethereum co-founder Vitalik Buterin told attendees at the OKX Now 2026 Summit in Singapore on October 6, 2026 that AI agents could handle most of the network's transactional activity within two years, with traditional on-chain applications receding behind agent workflows.

Speaking at the conference, Buterin argued that AI could become the new user interface for on-chain activity. Instead of opening a dedicated app, connecting a wallet and signing transactions, a user would instruct an AI agent to carry out the task directly on the blockchain. Buterin's framing was conditional: "what may happen, not what will," according to a write-up of the session.

To illustrate, Buterin described updating his own Ethereum Name Service (ENS) records in roughly five minutes by directing a local AI agent, with no conventional user interface involved. ENS maps readable names to Ethereum addresses; record updates typically require a wallet signature inside a dedicated application. The anecdote suggests that, for sophisticated users, the app layer is already optional.

What does the "cryptographic world computer" model change?

Buterin tied the agent thesis to a network transition he describes as a "cryptographic world computer" model. Under that design, one party executes a task and the rest of the network checks a compact proof that the work was done correctly, rather than re-running the computation. The approach reduces redundant execution across validators and reframes Ethereum's value proposition around verifiability.

Buterin has been linking Ethereum's roadmap to AI since early 2026, including remarks on trustless and private AI interactions, agents that participate in economic exchanges, and AI-driven verification of transactions. He also pointed to the Hegotá upgrade, planned for 2027, calling it potentially Ethereum's "last fork that developers from the network's 2015 era would still recognize." That timeline lines up loosely with his two-year horizon for AI-mediated activity.

What security risks did Buterin flag?

Buterin's talk was not purely optimistic. He named three categories of risk that intensify when agents hold signing authority over wallets:

• Privacy: pseudonymity on Ethereum could erode as AI systems aggregate user behavior
• Prompt injection: malicious instructions embedded in an AI's inputs can redirect the agent against its operator's intent
• Compromised agents: bots authorized to move funds become high-value targets, since attacking the agent is often cheaper than attacking the protocol

Buterin also referenced a separate incident that caused roughly $1.4 billion in losses by exploiting the user-interaction layer, the place where humans meet the chain. His point: replacing humans with agents does not by itself harden that layer; it relocates the attack surface.

What changes for wallets and protocol teams?

The operational consequence falls on wallet and application developers. If AI agents become the dominant interface, signing flows, transaction simulation and permission management have to be designed for machine clients, not only human users. Recovery paths, policy enforcement and key handling move from edge cases to core product features.

Buterin's personal example — updating his own ENS records through a chat with a local agent — signals where at least one influential voice believes Ethereum's user experience is heading. Whether the broader market follows depends on whether wallet and protocol teams ship agent-native primitives before the Hegotá fork activates in 2027.