Blockchain, Agents, and Robots: How the Cryptocurrency Industry Is Changing AI

When we first wrote about the potential for interaction between AI and blockchain projects in 2023, the sector was still in its formative stage. The potential and the main directions for convergence remained unclear.

In 2024, after the launch of Truth Terminal, followed by the first algorithmic fund ai16z, blockchain agents became the dominant narrative. Tokens of the largest platforms for launching them posted rapid growth, eclipsing so-called “AI memecoins” and the few technical projects that, at the time, operated at the intersection of artificial intelligence and blockchain. 

By spring 2025, the hype had faded. Most of the agents created proved ineffective, and their use cases were largely limited to social media posting and entertainment products. The sector’s total market capitalization shrank significantly, and user activity and project revenue fell — something clearly reflected in the metrics of the largest agent-launch platform, Virtuals.

Against this backdrop, traditional companies like Google, Amazon, and OpenAI have rolled out a range of solutions for building autonomous programs, including protocols (A2A and ACP), frameworks (AgentKit, Vertex AI Agents), and even platforms like Amazon Bedrock. 

This has allowed tech giants to shift the spotlight toward themselves, but it has not hollowed out the AI segment within the crypto industry, as tokenization has provided enough incentives to attract developers and fund their work.

According to Cookie.fun, the combined market capitalization of blockchain agents at the time of writing exceeds $5 billion. That is about a quarter of the total market — roughly the same share attributed to Near and ICP, which are not directly related to neural networks but are included in the relevant category on CoinGecko.

At the same time, not only the number of agents is growing, but their quality as well. For example, sophisticated automated trading systems have emerged. They make it possible to assess how effectively LLMs handle this kind of activity, as well as to copy a neural network’s trading strategy.

Development continues in the DeFAI segment as well, including through the emergence of new agents. In addition, a separate category of services is taking shape at the intersection of AI and prediction platforms, which gained popularity in the second half of 2025. 

However, a fundamentally new development vector is the shift from a collection of fragmented solutions to cohesive ecosystems, where agents can interact both with each other and with third-party services.

The key concept shaping the further development of autonomous assistants is agentic commerce. It involves AI algorithms executing deals and transactions on behalf of people and is part of the broader concept of an agentic economy.

According to Forbes, by 2030 the total value of transactions executed by AI could reach $30 trillion. The potential and importance of this direction are also underscored by recent announcements from major companies:

• in September 2025, Stripe announced the launch of ACP, enabling algorithms to make payments. OpenAI added support for this solution, making purchases possible directly within the ChatGPT interface
• in the same month, Google announced AP2 as an add-on to A2A, which sets the settlement rules between assistants.

Notably, Virtuals’ ACP protocol (not to be confused with Stripe’s counterpart) was announced back in February 2025. It sets a standard for agent-to-agent interactions and defines how blockchain operations are executed. Essentially, it is an analogue of centralized solutions, optimized for the crypto industry.

However, protocols for executing deals between algorithms are only part of the required infrastructure. Analysts at LongHash Ventures highlight three key components needed to build agentic commerce:

• Identification
• Transactions
• Training and personalization.

The last remains the most difficult to implement, but the rapid development of the first two already suggests that the foundations of a new economy are in place. Clear examples include automated shopping via ChatGPT or the Blormmy assistant, which can pay for your Amazon purchase using USDC from a linked wallet.

In the future, the number and capabilities of commercial algorithms will only continue to grow. And the crypto industry will provide the tech stack needed to make that happen.

In 2025, we are seeing parallel development of infrastructure for the agent economy by Big Tech companies and blockchain developers. While the former have deeper resources — in the first half of 2025 alone, investors poured around $700 million into projects tied to autonomous software — new solutions are also emerging in the crypto industry. 

The shared goal is to provide all the components needed to build sophisticated autonomous algorithms required for the new economy.

Identification

As the number of blockchain agents grows, a new problem has emerged — trust. And it shows up on two levels at once:

• for platforms and services, the question is whether a program can be granted access to a product or service; 
• for users, the issue is the assistant’s reputation and quality — how safe and effective it is, and whether it can be trusted with funds at all.

This trust crisis stems from how relatively easy it is to launch assistants and how simple it is to copy code from open repositories. That accessibility attracts inexperienced developers who may introduce technical vulnerabilities, as well as scammers who use agents for illicit activity.

The first step toward overcoming this limitation is identifiers — unique records or labels that make it possible to distinguish one agent from another. In Google A2A, this is handled via a so-called agent card, which provides data about the program. Virtuals addresses the problem through tokenization, linking a specific assistant to an NFT containing information about it, while Fetch.ai developed the Agent Name Service (ANAME) for the same purpose.

DNS for agents

Agent Name Service allows agents to be given human-readable ‘names’. Just as DNS translates an IP address into a website domain, ANAME converts a long blockchain address into a short, human-readable name. For example, alice.helper, instead of an address like f1a2b3c4…0.

This makes agents more recognisable and ensures their uniqueness, which is useful when the system contains many agents and reliable, understandable routing is required.

Пример доменного имени для агента. Данные: Fetch.ai.

Reputation systems are built around these identifiers. They let users rate the quality of an outcome, and projects approve or reject interaction requests. For example, a Virtuals agent profile includes not only the transactions it has executed, but also ratings and reviews from people who used its services.

Similar solutions exist on other platforms — uAgents from Fetch.ai, or Privado ID from SingularityNET. Among centralized projects, Visa’s TAP standard stands out. Algorithms that receive this “badge” are recognized by merchants as trusted. Something like a verification checkmark in Telegram or X. 

However, since each ecosystem or venue builds its own reputation system, evaluating solutions requires compatibility with multiple platforms at once.

A global fix could come from the ERC-8004 standard, introduced in August 2025 by a team that included developers from MetaMask, Google, and the Ethereum Foundation. It is essentially a universal “agent passport” in the form of an NFT. 

The token contains core data and user feedback, while compliance with the description must be confirmed by validators. The latter can be restaking services or, for example, oracles — depending on the nature of the data being verified. 

As of writing, the ERC-8004 registry contains more than 3,000 entries, and the standard itself has been integrated into a number of services, including agent launch platforms, specialized blockchains, and marketplaces. 

The solution is positioned as neutral, with no ties to any service or blockchain. In other words, you can register any agent, regardless of which framework it uses. However, it is obvious that only platforms operating on EVM-compatible networks will be able to integrate a reputation system based on ERC-8004.

If it sees broader adoption, the crypto market’s largest DeFi ecosystem will gain a comprehensive catalog of autonomous programs. Protocols will be able to use it as a “filter” for malicious algorithms, and users will be able to choose the highest-quality and most relevant product.

Interaction

Interaction between autonomous programs, as well as with third-party systems for executing deals, is one of the central challenges of the agent economy. Broadly, it can be divided into two levels:

• top level — coordination of algorithms and standardization of data exchange. It defines how different agents should exchange value and other information
• base level — the value transfer mechanism. That is, how an assistant can make a payment and receive the required service or product.

The leading top-layer protocol in the crypto industry remains the already mentioned ACP by Virtuals. A similar protocol — AITP — is also being developed by the Near team. They compete with centralized counterparts like A2A and ACP from Stripe.

The problem with these solutions is fragmentation, because programs built on different standards cannot interact with each other. This creates closed, competing ecosystems instead of a cohesive layer for the digital economy.

As for the base layer, the latest big news was the launch of the x402 protocol, introduced jointly by Cloudflare and Coinbase in September 2025. We covered it in more detail in an overview article.

People have been saying since the first mass-market AI products appeared that cryptocurrency would provide the financial infrastructure for agents. And it really happened — many algorithms, including Truth Terminal, got wallets and the ability to transact. However, most merchants still do not accept payments in digital assets, and those that do use processing solutions from intermediaries. 

While assistants can transfer tokens to another address, processing an invoice or logging in is too difficult for most programs. x402 removed this barrier.

The protocol has been integrated into the stacks of Virtuals, Google, and a number of other projects. The rollout of the new standard also benefited the USDC stablecoin as the “default” settlement currency, and the Coinbase crypto exchange, which performs transaction verification by default. 

Still, as of writing, about 2 million transactions totaling $615,000 have been processed through the protocol, and fewer than 1,000 merchants accept payments in this format, despite the ease of integration. This points both to a focus on small transactions and to relatively low adoption of the solution.

Another issue is that most applications built on the new standard are still in the early stages of development. Many remain vulnerable to attacks, creating additional risks for users. 

Addressing these shortcomings is a whole layer of work, with huge opportunities for new projects and entrepreneurial teams. 

Adaptation and Training

Adapting an agent to the needs of a specific user and revising its strategy based on accumulated experience is one of the toughest challenges for developers. Even without factoring in the general limitations of LLMs around the context window and the degradation of response quality as knowledge builds up, the main problem is data storage. 

Proprietary platforms — Google, OpenAI, Anthropic — provide developers with their own storage as part of the technical stack for building autonomous assistants. The recorded data can be managed, but this model still requires trusting the provider and, technically, opens access to the information for third parties. 

Some platforms, like Virtuals or ElizaOS, include long-term memory modules that store knowledge graphs and other information to adapt the agent. 

But these are hybrid systems — on-chain, only hashes or links to the primary data stores are available. Those backends can be either fully proprietary servers or decentralized networks like Chromia.

Some new projects are building fully distributed infrastructure. For example, Unibase provides decentralized “memory” for blockchain algorithms, including its own data availability (DA) layer and a protocol for coordinating and interacting with stored information. DeAgentAI follows a similar model. 

The EigenLayer ecosystem is particularly notable in this regard. The project effectively provides a full technical stack for blockchain agents, including:

• decentralized EigenCloud infrastructure for running software
• an interface for transparent interaction with the EigenAI model
• execution in a secure environment via EigenCompute
• storage of activity and request data in EigenDA. 

The key feature of this infrastructure is decentralization and transparency. The operation of a neural network launched on top of it can be tracked at virtually every stage, without needing to trust the provider or the developers. 

In practice, however, many blockchain agents are centralized, since developers can modify the code or memory at their discretion. At the same time, they may potentially lose out to solutions built on proprietary frameworks, because big tech companies can offer more storage space and flexible data management systems.

However, it is precisely user data — needed to tailor and personalize assistants — that could secure blockchain platforms a sustainable share of the agent economy market. 

Given the scale and advantages of centralized companies, the question arises — do blockchain assistants even stand a chance of surviving in such a competitive environment? To answer it, let’s go back to basics. 

Blockchain can still give autonomous algorithms two key advantages:

• tokenization
• decentralization

The first aspect is the ability to tokenize any program, making it possible to own it and receive a share of the profits from its activity. Virtuals offer a well-developed economic model for this. And while “virtual influencers” generate relatively modest revenue, commercial agents have more potential ways to monetize. 

Solutions from Google or OpenAI may operate an order of magnitude more efficiently, but only the developers will capture the revenue. In this model, users are relegated to passive consumers, so owning an assistant as a blockchain asset could appeal to retail investors. 

As of writing, Virtuals agents collectively earn around $50,000 per day, despite relatively limited functionality and waning interest in the segment. At the peak, the figure reached roughly $500,000. Part of these funds is distributed to token holders directly and via a buyback mechanism.

Even if you factor in that these are relatively small amounts, proprietary solutions from tech giants do not offer anything like this at all. That is why tokenization remains one of the key “moats” for blockchain agents. 

As for decentralization, its value increases as agents expand their scope. Their core — the LLM — will remain centralized for the foreseeable future, but for memory and personalization modules, as well as identity systems, blockchain provides a number of advantages:

• immutability of records
• censorship resistance
• protection and management of personal data

This is especially important in the era of agentic commerce, when a model can process payments, transfers, and purchases, tying them to contextual information from a user’s queries. Potentially, this is a massive trove of marketing and personal data that needs protection and governance. 

As an additional advantage, it is worth noting default support for digital assets amid the growing adoption of stablecoins as the primary instrument of the digital economy. They may be centralized and fully controlled by issuers, but one way or another they operate on decentralized networks, which makes them suitable for blockchain applications. 

That said, it is important to understand that stablecoins are no longer an “exclusive” feature of decentralized agents. Many payment providers, including Visa, Mastercard, Stripe, and others, process token transactions. This means any partner AI products will be able to use them, even fully centralized ones. 

The key question is whether developers of proprietary solutions can ensure proper protection of user data collected by their software. This is a difficult task, because the technical stacks of Google or Anthropic are open, which implies the existence of a huge number of independent builders. 

Decentralized frameworks can offer more than moderated “marketplaces” and privacy policies. Blockchain platforms have zero-knowledge proofs, as well as transparent ledgers.

And unlike in the traditional economy, in agentic commerce big tech does not yet have advantages in the form of protectionist regulation or a natural monopoly.

In this way, blockchain developers are following the same trajectory as large centralized companies, focusing on agent development and building the infrastructure for their operations. As a result, we are seeing two technical stacks evolve in parallel:

• proprietary, but high-performance and easy to scale
• open and transparent, but constrained by the specifics of blockchain

The first has greater potential for adoption thanks to the network effects of major platforms, while the second can provide privacy and open economic models. 

In a sense, the situation resembles the standoff between the crypto industry and the financial sector that lasted until 2024 and ended in the de facto merger of both “worlds.” But will the AI industry face the same fate? The question remains open.

In the next part, we will look at another sector related to AI on blockchain — DePAI, or decentralized robotics. According to some estimates, by 2030 this market could reach $126 billion. And even today, we are no longer talking about lab research or concepts, but about real, working projects.