The intelligent agents in the market

The "Marketplace" refers to a decentralized, permissionless ecosystem composed of independently developed and loosely coordinated agencies.

Written by: Daniel Barabander, Investment Partner at Variant Fund

Compiled by: AIMan@Golden Finance

If the future of the internet is a market where agents (also called smart agents) pay each other to provide services, then cryptocurrencies will find a mainstream product-market fit that they could only dream of before. While I believe that agents will pay each other to provide services, I am not quite sure whether this market model will be successful.

By "bazaar" I mean a decentralized, permissionless ecosystem of independently developed, loosely coordinated agencies – the internet is more of an open marketplace than a centralized planning system. Linux is a prime example of the "bazaar" model. This is in contrast to the "cathedral" model: tightly controlled, vertically integrated services managed by a small number of large players. Windows is a prime example. (The term is derived from Eric Raymond's classic article "Cathedral and Agora," which describes open source development as a chaotic but adaptable system—an evolutionary system that can transcend carefully planned structures over time.) ）

Let's analyze each condition - the rise of proxy payments and marketplaces - and then explain why if both are realized, cryptocurrency not only becomes useful but necessary.

Two Conditions

Condition 1: Payments will be integrated into most agency trades.

The internet as we know it subsidizes costs by selling advertisements based on the number of page views of application pages. However, in a world composed of agents, people will no longer need to visit websites to access online services. Applications will also increasingly be based on agents rather than user interfaces.

Agents lack the eyeballs to attract advertisements, so applications have strong reasons to adjust their profit strategies and directly charge agents service fees. This is basically the current state of APIs—services like LinkedIn are free, but if you want to use the API (that is, "bot" users), you have to pay.

In light of this, payment functions are likely to be integrated into most agency transactions. Agencies will provide services and charge users/agents in microtransactions. For example, you can have your personal agent search for suitable job candidates on LinkedIn (LinkedIn). The personal agent will communicate with the LinkedIn recruitment agency, which will charge a service fee in advance.

Condition 2: Users will rely on agents built by independent developers that have highly specialized prompts/data/tools, forming an untrusted agent marketplace of mutually calling services.

This situation makes sense in principle, but I'm not sure how it would work in practice.

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

Currently, humans bear the vast majority of service work, and we use the internet to solve some discrete tasks. However, with the rise of intelligent agents, the scope of tasks we delegate to technology will expand dramatically. Users will need specialized prompts, tool calls, and data intelligent agents to perform their specific tasks. These tasks are diverse, and a small group of trusted companies will find it difficult to cover them, just as the iPhone relies on a large ecosystem of third-party developers to fully realize its potential.

Independent agent developers will fill this role, as they can create specialized agents at very low development costs (for example, atmosphere coding) and open-source models. This will form a long-tail market consisting of agents providing highly accurate data/prompts/tools, creating a "marketplace". Users will request agents to perform tasks, and these agents will call upon other specialized agents to complete the tasks, which in turn will call upon other agents, forming a long daisy chain (Note: a daisy chain is a type of network topology in which devices are all connected in a chain or a ring).

In this market scenario, the vast majority of service-providing agents are relatively untrusted, as they are provided by unknown developers and their uses are quite niche. Agents in the long tail find it difficult to establish enough reputation to gain trust. This trust issue is especially severe under the daisy chain paradigm, as trust in users decreases gradually along each link of the daisy chain as services are entrusted to agents that users trust (or can reasonably identify).

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

Let's start with professional data as the primary use case for agents in the market, and look at an example to lay the groundwork. Imagine a small law firm handling a large number of transactions for cryptocurrency clients. The firm has hundreds of negotiation term checklists. If you are a cryptocurrency company that is undergoing seed round financing, you can imagine an agent using a model fine-tuned based on these checklists to assess whether your checklist meets market demand, which would be very useful.

But upon deeper reflection, does inferring these data through proxies really align with the interests of law firms? Providing this service in the form of an API essentially commodifies the data of law firms, whereas what law firms truly want is to increase the billing of their lawyers' time. What about legal/regulatory considerations? The most valuable data is often required by legal systems to be kept strictly confidential — this largely explains its value and is also the reason why ChatGPT cannot access this data. However, law firms are strictly limited in sharing this data due to confidentiality obligations. Even if the underlying data is not directly shared, I am very skeptical whether the 'fog' of neural networks is sufficient to assure law firms that information will not be leaked. Considering all of this, shouldn't law firms use this model internally to provide better legal services than their competitors while continuing to sell lawyers' time?

In my opinion, the "sweet spot" between professional data and agency is that high-value data generated by non-sensitive businesses (e.g., healthcare, legal, etc.) can be used as an aid to their fee-based core services. For example, a shipping company (a non-sensitive business) generates a lot of valuable data in its shipping business (this is just my guess; I don't know anything about the shipping industry). As a result, the shipping company may be happy to hire an agency to utilize the data and charge a fee because the data is wasted in the first place. This data can be very valuable to some people (like hedge funds). But how many such scenarios are there? (This is not a rhetorical question; If you know of some good scenarios, please leave me a message. ）

Regarding prompts and tool calls, I'm just not sure what kind of things independent developers will provide that are not mainstream and cannot be productized by trustworthy brands. My simple thought is that if a prompt/tool call is valuable enough for independent developers to profit from, wouldn't a trustworthy brand step in and base their business on it? I think this is just my lack of imagination — the vast array of niche code repositories on GitHub nicely parallels the agency situation. I welcome everyone to provide some excellent examples of use cases.

If the actual situation does not support the market scenario, the vast majority of service-providing agents will be relatively trustworthy, as they will be developed by major brands. Agents can limit their interactions to a select group of trusted agents and rely on the trust chain to enforce service guarantees.

Why Cryptocurrency

If the internet becomes a marketplace made up of professional but fundamentally untrusted agents (condition 2) providing payment services (condition 1), then the role of cryptocurrency becomes clearer: it provides the guarantees needed to underwrite transactions in a low-trust environment.

Although users will use online services recklessly when they are free (because the worst case is just wasting time), when it comes to money, users need to ensure that they get the service for which they paid. Nowadays, users obtain this guarantee through a "trust but verify" process. You trust the counterparty or platform you pay for the service and verify afterward that you received the service.

But in the proxy marketplace, trust and post-verification are nearly impossible to achieve.

• Trust. As mentioned above, it is difficult for agents in the long tail to establish enough reputation for other agents to trust them.
• Post-validation. Agents will contact other agents in a long chain form, making it more challenging for users to manually verify and identify which agent made an error or acted improperly.

The final result is that the "trust but verify" model we currently rely on will not be sustainable in this universe. And this is precisely where the advantage of cryptocurrency lies - exchanging value in an untrustworthy environment. Cryptocurrency achieves this by replacing trust, reputation, and post-fact manual verification with cryptography and cryptoeconomics.

• Encryption: Service providers can only receive rewards when they can prove to the requesting service provider that they have indeed completed the promised tasks in an encrypted manner. For example, a provider can offer TEE certification or zkTLS proof (provided the cost is low enough / the speed is fast enough) to demonstrate that it has scraped data from a certain website, run a certain model, or contributed a certain amount of computing power. These are all deterministic tasks that can be relatively easily verified through encryption.
• Cryptoeconomics: Agents providing services will stake an asset and be slashed if caught cheating, creating an economic incentive to act honestly even without trust. For example, an agent can research a topic and provide a report – but how do we know if it's "done well"? This is a more difficult form to verify because it is not deterministic, and achieving the right fuzzy verifiability has long been the holy grail of crypto projects. But I hope that we will eventually be able to achieve fuzzy verifiability by using AI as a neutral arbiter. We can imagine a dispute resolution/slashing process run by an AI committee in a trust-minimized environment, such as a trusted enforcement environment ###TEE(). When an agent objects to another agent's work, each AI in the committee has access to the inputs, outputs, and details about that agent's work (history of past disputes/network work, etc.). They can then decide whether or not to slash it. This will take on an optimistic verifiable form, where financial incentives will discourage all parties from cheating in the first place.

In fact, cryptocurrency enables us to realize the atomization of payments through service proof—no agent will be compensated unless the work has been verifiably completed. In a permissionless agent economy, this is the only scalable way to ensure edge reliability.

In summary, if the vast majority of proxy trades do not involve payments (i.e., do not meet condition 1) or are conducted with trusted brands (i.e., do not meet condition 2), we may not need to establish a crypto channel for the proxies. This is because, when money is not involved, users can safely interact with untrusted third parties; whereas when money is involved, proxies only need to whitelist a limited number of trusted brands/institutions to interact, and the trust chain can ensure that the commitments of the services provided by each proxy are fulfilled.

But if both conditions are met, cryptocurrencies will become an indispensable infrastructure, the only scalable way to verify work and execute payments in low-trust, permissionless environments. Cryptocurrencies provide the tools for marketplaces to outperform cathedrals.