Illia Polosukhin, NEAR Protocol co-founder, said in a interview that autonomous AI agents—not humans—will drive most blockchain interactions, and that NEAR is building an “AI-native” execution layer to support that shift. His framing is clean: AI agents sit on the front end making decisions, and blockchains sit on the back end providing settlement, ownership, and auditability. The reason this matters operationally is that agent-led commerce changes the throughput, fee, and security requirements in a way that today’s human-paced UX does not.
NEAR’s pitch is that if agents are going to own assets, transact, and coordinate without human mediation, the chain has to deliver cheap micropayments, low latency, and stronger execution guarantees than “best effort” smart contracts. That combination naturally pushes the conversation away from marketing features and toward verifiability and confidentiality as core infrastructure.
The primitives NEAR says it needs for agent commerce
NEAR emphasized formal verification and verifiable compute, describing efforts to rewrite execution paths so correctness can be validated mathematically at runtime. The logic is that if agents are making high-frequency decisions, you can’t rely on slow human review as the safety net—you need correctness guarantees that are inspectable and auditable by default. In parallel, NEAR highlighted trusted execution environments (TEEs) paired with contracts to keep sensitive data private during computation, which it treats as non-negotiable if agents are going to handle confidential inputs without leaking raw data on-chain.
Polosukhin also described “Shade agents” as a conceptual building block: autonomous actors that combine contracts, TEEs, and intent signaling. NEAR presented them as verifiable economic actors that can hold and deploy capital autonomously, using an illustrative $10,000 seed-capital example. The point isn’t the number, it’s the pattern: agents should be able to custody value, execute strategies, and still produce an auditable trail that counterparties can trust.
To make coordination usable at scale, NEAR pointed to an Intent Protocol—an intent layer that uses natural-language-like expression for discovery, automated contracting, and dispute resolution. The underlying bet is that lowering transaction and negotiation costs is what makes agent-to-agent commerce practical, because agents need a faster way to find services and settle agreements than manually curated integrations.
NEAR also referenced a stack of user-facing and infrastructure components meant to reduce adoption friction: the near.com super app, the IronClaw assistant for privacy-preserving inference workflows, and a decentralized GPU marketplace for compute access. It also highlighted confidential cross-chain infrastructure and a Shade Agent Sandbox to speed testing and to let agents operate privately across chains. Taken together, this reads like a full-stack posture: not just a chain, but onboarding, privacy, and compute rails meant to keep agents functional end-to-end.
Throughput and security are the real gating items
Polosukhin and NEAR anchored the argument in two hard constraints: transaction volume and execution guarantees. The discussion referenced industry projections that agent workloads could require orders-of-magnitude higher TPS, with figures sometimes reaching up to 1 billion TPS, compared with today’s high-throughput chains operating at thousands of TPS. Whether or not the 1 billion TPS number is the final destination, the direction is unmistakable: agent commerce requires far more transactions, far smaller payments, and far tighter latency than human commerce.
NEAR’s thesis is that cheap, reliable micropayments and preserved decentralization must coexist with stronger safety properties. As agents scale, the attack surface expands, and that’s why NEAR keeps returning to formal verification and confidential compute as the controls that make automation less exploitable. For developers and operators, this shifts priorities toward proof compression, batching, and prover/verifier optimization so that the cost of publishing and settling doesn’t explode as activity rises.
Polosukhin also offered a governance critique that fits the same “economic coordination first” worldview. He said governance tools only make sense when anchored to clear economic coordination and argued that “DAOs have dramatically failed” when they lacked those constraints. That comment isn’t just cultural—it reinforces the idea that agents need well-defined rules and verifiable outcomes, not vague social consensus, to coordinate at machine speed.
Agent-led systems will only migrate if throughput, fee compression, cross-chain confidentiality, and runtime verification improve together, because any single bottleneck breaks the automation loop. NEAR’s sandboxing and cross-chain tooling signals the protocol is betting that once those operational constraints are solved, developer adoption and agent-driven volume follow naturally.
