"Market" transcends "Cathedral", how does cryptocurrency become the cornerstone of trust in the AI agency economy?

Compiled by: Tim,

If the future internet evolves into a marketplace where AI agents pay each other for services, then to some extent, the mainstream products and market that cryptocurrencies will achieve will align with this scenario, which we could only dream of happening before. While I am confident that there will be service payments between AI agents, I still have reservations about whether the marketplace model will prevail.

By "marketplace", I mean a decentralized, permissionless ecosystem of independently developed, loosely coordinated agents. Such an internet is more like an open market than a centrally planned system. The most typical case of "winning" is Linux. In contrast to this is the "Cathedral" model: a vertically integrated, tightly knit service system dominated by a handful of giants, typified by Windows. (The term is derived from Eric Raymond's classic article "The Cathedral and the Bazaar," which describes open source development as seemingly chaotic but adaptable.) It is an evolutionary system that is capable of transcending elaborate systems over time. ）

Let us analyze one by one the two prerequisites for realizing this vision, namely the popularization of smart agent payments and the rise of the marketplace economy. Then explain why when both become a reality, cryptocurrencies will not only have practicality but will also become an indispensable presence.

Condition 1: Payments will be integrated into most agency trades

The internet as we know it relies on a cost-subsidy model that places advertisements based on the human browsing volume of application pages. However, in a world dominated by intelligent agents, humans will no longer need to personally visit websites to obtain online services. Applications will increasingly shift towards an architecture based on intelligent agents, rather than the traditional user interface model.

Intelligent agents do not have "eyeballs" (i.e., user attention) available for advertising sales, so applications urgently need to transform their profit strategies to charge intelligent agents directly for service fees. This is essentially similar to the current business model of APIs. Taking LinkedIn as an example, while its basic services are free, one must pay a corresponding fee to access its API (i.e., the "bot" user interface).

It seems that payment systems are likely to be integrated into most agent transactions. When providing services, agents will charge fees to users or other agents in the form of microtransactions. For example, you might ask your personal agent to look for excellent job candidates on LinkedIn, at which point your personal agent will interact with the LinkedIn recruitment agent, which will charge the corresponding service fee in advance.

Condition 2: Users will rely on agents built by independent developers, equipped with highly specialized prompts, data, and tools. These agents form a "marketplace" through mutual service calls, but there is no trust relationship among the agents in this marketplace.

This condition makes sense theoretically, but I'm not sure how it will operate in practice.

The following are the reasons for the formation of the market model:

Currently, humans undertake the vast majority of service work, solving specific tasks via the internet. However, with the rise of intelligent agents, the range of tasks that technology can take over will expand exponentially. Users will need intelligent agents with dedicated prompt instructions, tool invocation capabilities, and data support to complete specific tasks. The diversity of these task sets will far exceed the coverage capabilities of a few trusted companies, just as the iPhone must rely on a massive ecosystem of third-party developers to unleash its full potential.

Independent developers will take on this role, gaining the ability to create specialized intelligent agents through the combination of extremely low development costs (such as Video Coding) and open-source models. This will give rise to a long-tail market composed of a vast array of niche agents, forming a marketplace-like ecosystem. When users request agents to perform tasks, these agents will call upon other agents with specific professional capabilities to collaborate. The called agents will further invoke more vertical agents, thus forming a progressively layered collaborative network.

In this marketplace scenario, the vast majority of service providers are relatively untrusted among each other, as these agents are provided by unknown developers and serve rather niche purposes. Agents at the long tail end will find it difficult to establish enough reputation to earn trust recognition. This trust issue will be particularly prominent under the chrysanthemum chain model, where services are delegated in layers; as the distance between service agents and the initially trusted agent (or one that the user can reasonably recognize) increases, the user's trust will gradually diminish at each delegation stage.

However, there are still many unresolved issues when considering how to implement this in practice:

We will start from professional data as a main application scenario for intelligent agents in the market, deepening our understanding through specific cases. Suppose there is a small law firm that handles a large number of transactions for cryptocurrency clients, which has accumulated hundreds of negotiated term sheets. If you are a cryptocurrency company undergoing seed round financing, you can envision a scenario where an intelligent agent fine-tuned based on these term sheets can effectively assess whether your financing terms meet market standards, which would have significant practical value.

But we need to think deeper: does it really serve the interests of law firms to provide reasoning services for such data through intelligent agents?

Opening this service to the public in the form of an API essentially commercializes the proprietary data of law firms, while the true business demand of the firms is to obtain premium returns through the professional service time of lawyers. From a legal regulatory perspective, high-value legal data is often subject to strict confidentiality obligations, which is the core of its commercial value and also a significant reason why public models like ChatGPT cannot access such data. Even if neural networks possess the characteristic of "information fogging", under the confidentiality obligations between lawyers and clients, is the unexplainability of algorithmic black boxes sufficient for law firms to be confident that sensitive information will not be leaked? This poses significant compliance risks.

Considering all aspects, the more optimal strategy for law firms may be to internally deploy AI models to enhance the accuracy and efficiency of legal services, build differentiated competitive advantages in the professional services sector, and continuously use the intellectual capital of lawyers as the core profit model, rather than taking risks to monetize data assets.

In my opinion, the "best application scenarios" for professional data and intelligent agents should meet three conditions:

1. Data has high commercial value
2. From non-sensitive industries (non-medical/legal, etc.)
3. "Data by-products" that are not part of the main business.

Taking shipping companies as an example (a non-sensitive industry), the data generated during their logistics transportation process, such as vessel positioning, freight volume, and port turnover ("data waste" outside of core business), may have predictive market trend value for commodity hedge funds. The key to monetizing this type of data lies in: the marginal cost of data acquisition approaching zero and not involving core business secrets. Similar scenarios may exist in areas such as: retail customer flow heat maps (commercial real estate valuation), regional electricity usage data from power grid companies (industrial production index forecasting), and user viewing behavior data from film and television platforms (cultural trend analysis).

Currently known typical cases include airlines selling on-time performance data to travel platforms and credit card institutions selling regional consumption trend reports to retailers.

Regarding prompts and tool calls, I'm not quite sure what value independent developers can provide that hasn't been productized by mainstream brands. My simple logic is: if a combination of a prompt and a tool call is valuable enough to allow independent developers to profit, wouldn't trusted big brands directly enter the market to commercialize it?

This may stem from my lack of imagination; the long-tail distribution of niche code repositories on GitHub provides a good analogy for the agent ecosystem. Feel free to share specific examples.

If the real-world conditions do not support the marketplace model, then the vast majority of service-providing agents will have relative credibility, as they will be developed by well-known brands. These agents can limit the scope of interactions to a selected set of trusted agents, enforcing service guarantees through a trust chain mechanism.

Why is cryptocurrency indispensable?

If the internet becomes a marketplace composed of specialized but fundamentally untrustworthy agents (condition 2), who are compensated for their services (condition 1), then the role of cryptocurrency will become much clearer: it provides the necessary trust assurance to support transactions in a low-trust environment.

When users utilize free online services, they engage without hesitation (as the worst outcome is simply wasting time), but when it comes to monetary transactions, users strongly demand the certainty of "payment for value." Currently, users achieve this assurance through a "trust first, verify later" process, trusting the counterparty or service platform during payment, and verifying performance retrospectively once the service is completed.

However, in a market composed of numerous agents, trust and post-verification will be much more difficult to achieve than in other scenarios.

Trust. As mentioned earlier, agents in a long-tail distribution will find it difficult to accumulate enough credibility to gain the trust of other agents.

Post-verification. Agents will call each other in a long chain structure, making it significantly more difficult for users to manually check the work and identify which agent is negligent or acting improperly.

The key point is that the "trust but verify" model we currently rely on will be unsustainable in this (technological) ecosystem. This is precisely the area where cryptographic technology excels, as it enables value exchange in an environment lacking trust. Cryptographic technology replaces the reliance on trust, reputation systems, and post-facto manual verification in traditional models through the dual guarantees of cryptographic verification mechanisms and cryptoeconomic incentive mechanisms.

Cryptographic Verification: The agent executing the service can only receive compensation after being able to provide cryptographic proof to the requesting service's agent, confirming that it has completed the promised tasks. For example, the agent can prove via Trusted Execution Environment (TEE) proof or Zero-Knowledge Transport Layer Security (zkTLS) proof (provided we can achieve such verification at a sufficiently low cost or sufficiently fast speed) that it has indeed crawled data from the designated website, executed specific models, or contributed a specific amount of computing resources. Such tasks are deterministic in nature and can be relatively easily verified through cryptographic techniques.

Token Economics: Agents providing execution services need to stake certain assets, and if they are found to be cheating, they will be penalized and forfeited. This mechanism ensures honest behavior through economic incentives, even in a trustless environment. For example, an agent may research a topic and submit a report, but how do we determine if they have "performed excellently"? This is a more complex form of verifiability, as it is not deterministic, and achieving precise fuzzy verifiability has long been the ultimate goal of crypto projects.

But I believe that by using AI as a neutral arbitrator, we finally have a chance to achieve fuzzy verifiability. We can envision an AI committee running dispute resolution and forfeiture processes in trust-minimized environments such as trusted execution environments. When one agent challenges the work of another agent, each AI in the committee will receive the input data, output results, and relevant background information of that agent (including its historical dispute records and past work on the network). They can then determine whether to impose a forfeiture. This will create an optimistic verification mechanism that fundamentally prevents cheating by the participants through economic incentives.

From a practical perspective, cryptocurrencies enable us to achieve the atomicity of payments through service proofs, meaning that all work must be verified before the AI agents can receive compensation. In an agent economy without permissioned access, this is the only scalable solution that can provide reliable guarantees at the edge of the network.

In summary, if the vast majority of agency transactions do not involve the payment of funds (i.e., do not meet condition 1) or are conducted with trusted brands (i.e., do not meet condition 2), then we may not need to establish cryptocurrency payment channels for agents. This is because when funds are secure, users do not mind interacting with non-trusted parties; whereas when funds are involved, agents should limit the interactive parties to a small whitelist of trusted brands and institutions, and ensure the fulfillment of the service commitments provided by each agent through a trust chain.

But if both of these conditions are met, cryptocurrencies will become an indispensable infrastructure, as they are the only way to validate work and enforce payments on a large scale in a low-trust, permissionless environment. Cryptographic technology gives the "market" a competitive tool that surpasses the "cathedral."