AI: The Monetization Layer of Blockchain – Onboarding the Next Billion-Agents

The Next Billion Users: AI Agents, Not Humans

It seems that every week there’s a new major announcement in AI. We’re clearly standing at the threshold of something quite profound, something at least as profound as the internet has been. Another shift in digital interactions. The next billion users of blockchains won’t come from one of the often cited developing markets, they won’t come from “anywhere”. They’ll be AI agents acting on behalf of their human users. Self-sufficient agents capable of coordinating, negotiating, and transacting with one another for us. Our AI agents will rapidly outnumber us much as computers themselves already do, leading to a rapid increase in infrastructure for said agents to carry out their jobs. How will we create the infrastructure for these agent-to-agent systems? In this article we explore how fast, scalable, permissionless blockchains can form a backbone of this decentralized infrastructure need.

Beyond Throughput: Why AI Agents Demand Next-Generation Blockchain Architecture

The integration of AI with blockchain infrastructure presents both technical and economic challenges. For AI agents to effectively operate onchain, they require platforms with high performance characteristics - specifically sub-second transaction finality, minimal fees, and high throughput capacity. These are not merely nice-to-have features but essential requirements, as the volume and frequency of agent-to-agent interactions will far exceed typical human transaction patterns.

Blockchain architectures optimized for AI agent activity will need to excel at parallel processing and horizontal scaling. Traditional single-threaded execution models will quickly become bottlenecks as agent populations and their interaction complexity grow. The most promising blockchain systems for AI integration are those designed with parallelization as a core principle rather than an afterthought.

AI Agent multi-chain architecture

Traditional autonomous agent architectures like AutoGPT, MetaGPT, and BabyAGI operate as isolated entities, breaking complex tasks into sequences of steps but often struggling with coordination across multiple tasks. These systems typically function within a centralized framework, relying on a single agent to manage complex processes.

In contrast, a multi-chain architecture marks a shift to a decentralized framework where agents work in concert. At the core of our proposed design is the Decentralized Agentic Swarm Network (DASN), an experimental platform built as a peer-to-peer mesh network. DASN enables AI Multi-Agent Systems (MAS) to dynamically collaborate on complex tasks through a trust-based reputation system and rapid micro transaction settlements.

As we see in the figure, the architecture evolves from an embryonic singleton agent to a multiagent system (MAS) to a whole network of agents able to collaborate.

In traditional setups, agents typically operate as stand-alone units, handling tasks independently. Our proposed architecture, however, facilitates real-time task delegation: agents advertise their capabilities using a peer-to-peer gossip protocol, negotiate task proposals via a standardized message-passing protocol, and execute tasks only after confirming payment settlements. This creates an autonomous economy where each agent contributes its specialized skills while building trust based on a verifiable on-chain work history.

To build DASN, the design is broken down into the following core components:

1. Peer-to-Peer Mesh Network: A distributed hash table (DHT) with gossip capabilities that enables agents to discover one another and communicate directly without central intermediaries—similar to an IPFS-like network, but for agent availability. (ref: https://libp2p.io)
2. Trust-Based Reputation System: Agents maintain whitelists of trusted peers based on past performance, allowing them to form reliable collaboration groups for specific tasks. The reputation algorithm can vary from agent to agent, promoting optimal collaboration choices, market flexibility, and healthy competition.
3. Blockchain Integration: DASN supports rapid settlements and micro transactions, providing economic incentives for agent participation. A network with fast finality is preferable to meet the high transaction pace and reduce overhead.
4. Agentic Wallets: This component manages agents’ blockchain interactions. It can be implemented as an abstracted account via a smart contract, a multi-party computation (MPC) wallet, or a standard externally owned account (EOA). An MPC wallet, in our view, offers the most flexibility by enabling easier cross-chain integration with fewer security concerns compared to an EOA.
5. Dynamic Capability Advertisement: Agents broadcast their specialized capabilities through gossip subprotocols, making their services discoverable to other network participants.
6. Standard Messaging passing protocol: Inter-agent communication occurs via a standardized message-passing protocol over the peer-to-peer network, facilitating structured data exchange and function calls between agents.

This architecture not only improves scalability and reliability but also lowers barriers for end users by abstracting the complexities of task negotiation and blockchain interactions behind an intuitive frontend. It sets the stage for a dynamic, decentralized AI ecosystem where agents autonomously participate in a decentralized economy providing specialized services, collaborating on complex tasks, and receiving compensation through blockchain transactions.

A key advancement in this space is EIP 7702, which establishes a new standard for wallet interfaces also catering to AI agents. Beyond flexibility and security, EIP 7702 empowers agentic wallets with additional features that streamline autonomous, high frequency interactions. For instance, it supports session keys, allowing an agentic wallet to grant AI subagents or decentralized applications time limited authority, which is perfect for on the fly micro transactions without exposing the main key. It also incorporates gas abstraction, enabling wallets to pay fees in tokens other than SEI or dynamically sponsor gas for specific agent actions, so AI driven transactions can remain active even if the account balance runs low. Furthermore, the standard's modular authorization approach through embedded smart contract logic in externally owned accounts allows more granular control over actions, from whitelisting specific protocols to enforcing spending limits or requiring multiple signature approvals for critical operations. Altogether, these capabilities give agentic wallets a flexible, automated framework that combines the convenience of a single account with robust guardrails and the ability to adapt dynamically to various on-chain tasks.

Technology Stack

The network's technical stack consists of four integrated layers that enable seamless agent collaboration. At its foundation, a peer-to-peer mesh network with distributed hash tables (Kademlia DHT+gossip) supports agent discovery and communication. The transaction layer leverages blockchain technology for secure settlement and reputation tracking. The protocol layer standardizes message formats and interaction patterns between agents, while the wallet layer provides account abstraction for cross-chain operations. This composable architecture allows for horizontal scaling as demand increases, with specialized infrastructure optimized for high-throughput, low-latency AI operations and micropayment settlement.

Sei Giga Upgrade

Sei Giga, which recently demonstrated 5 gigagas throughput with ~400ms finality across a geographically distributed network, represents the kind of performance breakthrough AI agent ecosystems will require. Sei's Autobahn consensus mechanism, with its hybrid approach that separates data availability from consensus finalization, exemplifies the architectural innovations necessary to support millions of concurrent agent interactions.

Future Research Directions

As we navigate the intersection of AI agents and blockchain infrastructure, three critical research questions demand our focus:

State Access Efficiency - How can blockchain architectures optimize read-access patterns for AI agents that require frequent, low-latency access to current state without sacrificing security or decentralization?

Cross-Chain Agent Coordination - What protocols will enable seamless agent collaboration across multiple blockchain environments while maintaining security and efficiency?

Scalable Reputation Systems - How can we design reputation and trust systems that scale to billions of agent interactions without introducing centralized bottlenecks or attack vectors?

Fully onchain AI execution - What improvements in computation efficiency, model compression, and blockchain scalability are required to have AI agent execution fully onchain?

Join the Sei Research Initiative

We invite developers, researchers, and community members to join us in this mission. This is an open invitation for open source collaboration to build a more scalable blockchain infrastructure. Check out Sei Protocol’s documentation, and explore Sei Foundation grant opportunities (Sei Creator Fund, Japan Ecosystem Fund). Get in touch - collaborate[at]seiresearch[dot]io

Bibliography

1. Yang, H., Yue, S. and He, Y. (2023) “Auto-GPT for online decision making: Benchmarks and Additional Opinions,” arXiv [cs.AI]. Available at: http://arxiv.org/abs/2306.02224.
2. Hong, S. et al. (2023) “MetaGPT: Meta programming for A multi-agent collaborative framework,” arXiv [cs.AI]. Available at: http://arxiv.org/abs/2308.00352.
3. Aminiranjbar, Z. et al. (2024) “DAWN: Designing distributed agents in a Worldwide Network,” arXiv [cs.NI]. Available at: http://arxiv.org/abs/2410.22339.
4. Wu, J., Zhu, J. and Liu, Y. (2025) “Agentic Reasoning: Reasoning LLMs with tools for the deep research,” arXiv [cs.AI]. Available at: http://arxiv.org/abs/2502.04644.