Coinbase Venture: Overview of the entire track and business model

Written by Jonathan King Translated by TechFlow

Abstract

The fusion of cryptography and artificial intelligence is spawning a thriving on-chain AI economy, an ecosystem of blockchain applications and services driven by autonomous AI agents. Decentralized AI projects have seen significant funding and rapid growth over the past 18 months, but we believe on-chain AI is rapidly emerging, marking the next wave of innovation in this intersectional space. The importance of on-chain AI is that it expands the crypto space to potentially billions of AI-powered participants. Each autonomous AI agent is like a new “user” of the blockchain, capable of running 24/7 and making complex decisions, significantly driving on-chain activity and growth.

By investing in on-chain AI, Coinbase Ventures is supporting the builders of this future agent-based economy, paving the way towards a new “Agentic Web”. Coinbase Ventures portfolio companies mentioned for the first time in the following articles are marked with an asterisk (*).

In October 2024, Coinbase Ventures released a theoretical framework on the integration of encryption and AI, which pointed out that blockchain and AI have complementary advantages - blockchain provides decentralization, censorship resistance, verifiability and user ownership, while AI brings powerful data processing, reasoning and automation capabilities. We believe this synergy can revolutionize the way humans and machines interact in the digital economy, ultimately giving rise to an “agentized web” where AI agents operate on crypto infrastructure, driving significant economic activity and growth.

A key distinction is between decentralized AI and on-chain AI. Decentralized AI (“Crypto → AI”) refers to building general AI infrastructure that inherits the openness and peer-to-peer nature of blockchain networks. This includes efforts to democratize access to computing resources, data, models, and training, and to prevent AI development from being monopolized by a few large companies. These decentralized AI resources also power on-chain AI (“AI → Crypto”) — an ecosystem of applications and services that embed AI into new and old blockchain use cases (e.g. trading agents, on-chain portfolio managers, DeFi abstractions, etc.). While decentralized AI projects have seen significant funding and growth over the past 18 months, we believe on-chain AI is rapidly emerging, marking the next wave of innovation in this intersectional space.

Introduction to On-Chain AI

In the past year, we have witnessed an AI agent (such as Truth Terminal) equipped with a self-hosted wallet, creating an Internet-native religion, and launching a meme coin with a market value of more than $950 million, becoming the first AI agent "millionaire". According to cookie.fun , there are currently approximately 1,600 AI agents with a total market value of over $11 billion. Overall, we’re seeing AI agents (and associated “agent tokens”) quickly take over social channels, some with real utility, and moving on-chain AI from concept to thriving reality. In particular, the following three interrelated concepts are gaining attention: on-chain AI agents, on-chain AI applications, and agentized commerce.

On-chain AI agents are autonomous programs (driven by AI models) that are capable of performing on-chain operations. Think of an AI agent as a smart software robot with a crypto wallet — it can hold tokens, interact with smart contracts, trade assets, and even vote in DAOs, all based on its programming and goals. Unlike the isolated AI chatbots we commonly see on social platforms today, these agents are able to learn, reason, and act in an on-chain economy.

On-chain AI applications are blockchain applications that integrate AI into core functions. For example, AI can be embedded in DeFi protocols to optimize returns, in games to control NPC behavior, or in decentralized social networks or consumer applications to achieve high personalization of user content. While we’ll explore these examples later, the key point is that these applications are designed to seamlessly blur the line between blockchain and AI-driven logic.

Agency commerce is an emerging business model in which AI agents conduct transactions (including with humans) through blockchain. It’s a paradigm shift from manual, search-based transactions to a more automated, intent-driven and personalized transaction experience. Agents will become shoppers, negotiators, and service providers, completing transactions at the speed of software while aligning with human intent. Blockchain provides these agents with identities, wallets, stablecoins as payment currencies, and a smart contract framework for programmable transactions.

The importance of on-chain AI lies in expanding the crypto space to potentially billions of AI-driven participants. Each autonomous AI agent is like a new “user” of the blockchain, capable of running 24/7 and making complex decisions, laying the foundation for significant on-chain activity and growth. Next, let’s dive deeper into the burgeoning on-chain AI ecosystem and learn about its building blocks (new infrastructure services and on-chain agent types), emerging on-chain AI applications, and how business itself may be reshaped.

Agents

On-chain AI agents are the core of the “Agentic Web”. These are AI-driven entities capable of sensing, deciding, and acting in the on-chain economy. To understand their rise, we need to break down the infrastructure required to implement on-chain proxies and explore the types of proxies that are currently emerging.

Agent Infrastructure and Services

Building a powerful on-chain AI agent is very complex - it requires a whole new set of services and tools that are based on decentralized AI (DeAI) infrastructure resources (such as computing, data, models, intelligence, etc.) to support an open ecosystem of autonomous agents. These services make it easier to create, deploy, discover, and operate autonomous on-chain agents by abstracting complexity and providing reusable components. Below are the key emerging categories in proxy infrastructure and their role in the on-chain AI technology stack.

Trusted Execution Environments (TEEs)

In order to truly achieve autonomous and secure operation, on-chain AI agents require an execution environment that is tamper-proof, verifiable, and independent of any centralized party. A Trusted Execution Environment (such as Intel SGX or decentralized alternatives like Eternis*, Fleek*, or Phala Network) provides a hardware-secure “enclave” where an agent’s code and data can be processed confidentially, even from the agent creators themselves. Agents running in TEEs are protected from outside interference and can generate cryptographic proof that their behavior conforms to programmed instructions. As the proxy economy expands, embedding sovereignty into the infrastructure layer will be critical to earning user trust and enabling a fully autonomous proxy ecosystem.

Agent Frameworks & Tools

Agent frameworks (such as ElizaOS, G.A.M.E. by Virtuals, RIG, Heurist, REI) are development environments and libraries for building AI agents without developers having to start from scratch. These frameworks provide the architecture for the agent’s “core brain” — responsible for memory, decision-making, responding to cues, and task execution. On-chain agent toolkits (such as Coinbase AgentKit and SendAI) pre-package these frameworks for specific use cases and connect agents to smart contracts, wallets, payment channels, and on-chain data. By using these frameworks and tools, developers can quickly create powerful agents that have built-in support for advanced multi-platform interactions, long-term memory, and on-chain connectivity.

Agent Launchpads

Platforms in this category help create, launch, manage and/or monetize AI agents by packaging them as on-chain entities (usually with their own tokens). For example, agent launch platforms (e.g., Virtuals, auto.fun, ARC) allow creators to deploy new agent instances and build community or financial support around them. Through token or fee-aligned incentives, these launchpads enable proxy developers to maintain and scale their on-chain proxies as independent projects or businesses.

Multi-agent Coordination

Not all problems can be optimally solved by a single agent. Multi-agent coordination protocols (such as Virtuals ACP, Questflow, Theoriq) can coordinate multiple AI agents (i.e., "agent swarms") to work together to complete complex tasks. For example, one agent might be responsible for data collection, while another is responsible for evaluating the results, all of which are overseen by an on-chain coordinating agent. This “swarm” approach is able to surpass the capabilities of a single agent by leveraging specialization and parallel processing. By enabling cooperation between agents, multi-agent coordination platforms can extend the scope of on-chain AI automation, from multi-step workflows to entire autonomous organizations.

Model Context Protocols (MCP)

Model Context Protocols are a key service at the intersection of AI agents and external data, originally created by Anthropic. These protocols help standardize how on-chain agents obtain relevant context, knowledge, or tools from external sources. Rather than creating custom integrations for each data source or smart contract, agents integrated with the MCP standard can tap into any compatible context provider (whether it’s on-chain data, an off-chain database, or a web service) to retrieve the required information or tools. Decentralized MCPs such as Heurist and DeMCP provide agents with self-developed and open-source MCP services, enabling them to access mainstream large-scale language models in one stop, thereby enhancing the adaptability and capabilities of on-chain agents in practice.

AI App Stores

AI App Stores (such as Alchemist AI, ARC Ryzome) are platforms that serve as markets and discovery layers for on-chain agents, tools, and experiences. These app stores make it easy for developers to publish, monetize, and distribute agents or AI modules, while allowing users to browse, summon, or customize agents through familiar interfaces. These app stores are not only distribution hubs, but also coordination interfaces for the broader on-chain AI economy, facilitating interoperability between agents, tools, and protocols. As the number of on-chain agents and AI-native applications grows, these platforms may become important ecosystems — curating experiences, guiding users, and capturing a portion of the value that flows through agent interactions.

Agent Types

With the rapid development of agent infrastructure and service layers, we believe that on-chain AI agents can currently be roughly divided into the following categories:

Trading / DeFi Agents

These agents focus on financial operations, such as executing trades (such as Bankr*, Cliza), providing liquidity (such as BasisOS), optimizing yields (such as ARMA*, Mamo*), or performing arbitrage in DeFi. Additionally, they may participate in prediction markets (such as Billy Bets*) or even manage entire investment funds or portfolios (such as ai16z, aiXCB). These trading agents can react faster than humans, operate 24/7, and potentially make smarter decisions based on data, thereby making markets more efficient (or perhaps outperforming human traders in some ways).

Service Agents

Service agents provide practical services to users or protocols. For example, an agent may provide relevant market analysis research and insights (such as aiXBT, BitQuant*, Chaos AI*). Some proxies might handle DAO governance tasks — reading proposals, summarizing their contents, and even voting according to preset logic. Other service agents might audit smart contracts for vulnerabilities, or automatically generate new smart contract code based on natural language input (e.g., AgenTao, Kolwaii). In addition, there are business-related service agents (such as Byte AI), such as negotiating transactions or paying for goods on behalf of users. These agents are essentially “autonomous workers” in crypto, able to automate on-chain tasks that would normally require human labor or attention.

Entertainment Agents

These agents focus on interacting with users. In games, AI agents can act as NPCs (non-player characters) and interact naturally with players. Unlike traditional scripted game bots, these AI NPCs are able to learn and evolve, making the game more immersive. Beyond games, there are social agents: for example, AI influencers (like Luna) on platforms (like X or Farcaster*) that post content and interact with users, or AI agents that create artwork and IP based on community input (like Botto). In the future, you might follow an AI influencer on-chain that manages its own treasury (perhaps earning cryptocurrency by creating content on Zora8 or completing tasks for fans). There are also companion AI agents that can provide highly personalized interactions, some of which even have very detailed multimodal expressions and movements (such as Nectar AI).

While it is still early days, these categories demonstrate the broad possibilities of on-chain AI agents. From AI fund managers to AI virtual friends, on-chain agents can occupy multiple niches. What unifies them is that they are based on cryptographic technology, using cryptographic primitives as a "playground" and toolbox - holding assets, executing smart contract code, and taking advantage of the transparency and composability of decentralized networks.

Applications

Along with the rise of autonomous agents, we are also witnessing the rise of a wave of AI-driven on-chain applications. These applications and platforms embed AI into the user experience or core functionality. Here are some of the areas where on-chain AI applications are taking shape:

DeFi (“DeFAI”)

AI is entering the DeFi space in a variety of ways. One notable trend is AI-assisted trading and portfolio management. Instead of manually operating complex DeFi protocols, users can use the AI ​​interface to handle it for them. For example, HeyElsa is an AI-powered crypto assistant where users simply give its agent instructions for tasks (like “change X to Y”), and the agent will perform those actions across protocols. Protocols like Giza offer non-custodial agents that are able to monitor DeFi markets, identify yield optimization opportunities, and dynamically manage positions with real-time market awareness. We believe this AI-driven user experience marks the “Wealthfront moment in crypto,” where the on-chain AI agent acts as a robo-advisor designed specifically for DeFi, effectively becoming a personal crypto portfolio manager available to everyone.

Gaming & Agentic Metaverses

Games are a natural testing ground for AI agents. When combined with real asset ownership on the chain, the concept of an agentic metaverse is formed. These are game worlds or virtual environments populated by AI agents together with other agents or human players to create richer and more dynamic game content. These agents can be friendly NPCs (non-player characters), autonomous opponents, or even AI avatars controlled by other players. For example, Youmio is building an autonomous world where AI agents can learn, play, and entertain themselves in real time, creating a never-ending on-chain simulation. Additionally, companies like Farcade* are building an AI-powered on-chain game studio where anyone can “impromptly code” and distribute on-chain games via natural language prompts.

Consumer AI

AI is revolutionizing the consumer experience by making applications more personalized, interactive, and intelligent.

ChatGPT alternatives like Venice and FreedomGPT allow users to access powerful models in a privacy-preserving and censorship-resistant environment. In on-chain social networks, AI agents can act as influencers, curators, or creators — managing content streams, generating posts, participating in conversations, and even performing on-chain actions (like Clanker). In on-chain consumer applications like Zo, AI can help streamline the user onboarding process, recommend actions based on on-chain behavior, or negotiate on behalf of users in peer-to-peer marketplaces. Finally, AI companion agents like Nectar allow users to create and interact with agents that are capable of responding with nuanced multimodal expressions and actions — all of which can be verified on-chain. These proxyized experiences have the potential to significantly improve user experience in crypto, bringing it closer to what mainstream consumers expect.

Commerce

One ​​of the most profound impacts of on-chain AI is how it drives a whole new form of digital commerce — what Coinbase Ventures calls “Agentic Commerce.” This business model is driven by transactions between AI agents and humans or other agents. In such an economy, cryptocurrencies become the preferred payment method for machines and humans alike. The logic behind this is simple: autonomous AI agents operating around the world cannot access banks, but they can send and receive cryptocurrencies trustlessly on public blockchains. The borderless, programmable nature of cryptocurrencies makes them ideal for machine-to-machine payments, microtransactions, and automated contracts. For example, the Coinbase Developer Platform team recently launched x402, a new open source payment protocol that allows AI agents and applications to use crypto to pay for GPU computing, API access, digital content, and more. Additionally, startups like Payman* and Skyfire* are building infrastructure services that enable payment coordination between agents and humans or agents using stablecoins like USDC.

While agent-based commerce is still in its early stages, we believe it has the potential to automate and accelerate business transactions in unprecedented ways. Commerce could become machine-like and operate 24/7, with agents negotiating deals, executing contracts, and exchanging value in seconds. Importantly, humans set the goals and parameters, and the agent does the rest. The role of the blockchain is to provide a secure and interoperable “playground” for these agents to trade — with clear rules (smart contracts) and reliable currency (stablecoins).

Future Outlook

Looking to the future, the prospects of on-chain AI are full of potential, but its development will be carried out in stages. In the short term, we expect experiments with on-chain AI agents and AI-driven applications to continue. In the long term, we believe that cryptocurrency will become the de facto economic layer for AI, meaning that any advanced AI agent will use cryptocurrency to store value and settle transactions. As AI’s ability to write software and smart contract code continues to improve, the pace of innovation in the on-chain economy is likely to accelerate rapidly, bringing an influx of new applications and users.

However, there are still some challenges to overcome to realize this vision. Agent technology is still in its early stages, and some expectations may be ahead of reality. Current AI agents are still limited in reliability and range of capabilities, and it may take some time before they can safely handle open-ended tasks. At the same time, if a large number of agents conduct transactions at the same time, the scalability of the blockchain will be tested. In addition, the need for new trust and governance frameworks is urgent. While AI agents can greatly enhance the functionality of on-chain systems, they can also amplify security and trust issues if not properly governed.

From a value capture perspective, we believe that unleashing the economic potential of on-chain AI requires support from the following aspects: sound infrastructure to enhance agent intelligence (e.g., data networks and post-training models designed for on-chain agent use cases); services and tools for coordinating agent behavior (e.g., multi-agent coordination, decentralized MCP, agent identity/payment rails); channels for distributing agents to mainstream consumers (e.g., agent launch platforms, AI app stores, and consumer AI).

In short, the rise of on-chain AI represents a new frontier in machine-driven intelligence. From autonomous agents executing smart contracts to on-chain applications that adapt to user needs in real time, this movement has the potential to redefine how humans interact with machines. These are exciting times — and we at Coinbase Ventures and many in the crypto community believe this could lead to the next major leap in the evolution of the Internet, the advent of an “Agentic Web” that will drive a more autonomous and intelligent digital economy.

Thanks to Hoolie (Coinbase Ventures), Luca (Base), Lincoln (Coinbase), Vik (Coinbase), Daniel (Variant), Josh (Contango Digital), Anand (Canonical), Teng (Chain of Thought), and EtherMage (Virtuals) for their insightful feedback and discussions on this post.