Dragonfly Partner: Most agents will not engage in autonomous trading; how will crypto payments succeed?

Author: Robbie Petersen, Dragonfly Junior Partner

Translation: Gu Yu, ChainCatcher

Whenever a new narrative enters public discussion, mainstream arguments are often simplified into the most easily digestible form. Intuitively, when no one can empirically prove what will happen, provocative claims tend to be more rewarding than detailed analysis.

Recent discussions around “agent commerce” are no exception. The market has reached a consensus: the number of agents is rapidly increasing; agents need to conduct transactions; agents cannot hold bank accounts but can hold electronic wallets; card networks charge 2-3% fees; therefore, stablecoins will prevail.

This logical chain has flaws on many levels. Agents can hold bank accounts under an FBO (Financial Business Operator) structure. Additionally, the 2-3% fee reflects credit and fraud risks, which blockchain cannot solve.

However, the debate over “which payment method will win” actually stems from a fundamental premise often overlooked:

Will most agents actually conduct transactions?

The scale of agent economies will be enormous, but the proportion of agents actually transacting will not be as high.

Agent Economy Is More Like an Organizational Chart Than a Market

Fundamentally, artificial intelligence is an automation technology. It can perform certain tasks—such as searching, aggregating, and synthesizing—and do so more efficiently than humans. Agents are operational derivatives of AI. They don’t just return outputs; they perform actual actions.

The implicit assumption behind the entire agent commerce theory is: execution costs money. In other words, for most agent tasks, they need to spend funds to autonomously acquire external resources, pay for compute and data on a usage basis, and interact with other agents as independent economic entities.

This contradicts the actual application of agents.

Overall, agent deployment can be divided into two categories: business intelligence agents deployed by enterprises, and consumer agents that enhance our personal lives. For different reasons, neither type is likely to conduct autonomous transactions.

Commercial Agents Are an Inevitable Evolution of SaaS

A reasonable concept of commercial agents is that they are an inevitable evolution of SaaS. They do not enhance workflows; they replace existing workflows. Just as over 95% of software expenditure comes from enterprises and governments, over 95% of large-scale agent application scenarios are likely to be deployed within similar organizations.

This is the first subtlety overlooked by current mainstream agent business theories: the vast majority of agent needs are not for agents to book flights for consumers, but for top-down deployment within organizations. More importantly, agents automating tasks within closed organizations are fundamentally different from agents operating as independent economic entities.

Take sales agents as an example. They connect to CRM systems, research potential clients, craft personalized marketing copy, and schedule follow-ups. They do not spend money independently nor interact with external agents of other organizations. They simply perform a task—sales expansion—in a closed environment and automate it.

This scenario applies to nearly all organizational functions. Financial agents audit and verify expenses; accounting agents record journal entries, reconcile accounts, and prepare reports; legal agents review contracts and identify exceptions; coding agents write code.

In almost all cases, agents do not spend money themselves and are not granted spending authority. They are deployed in a top-down manner within a controlled organizational environment, with permission controls in place.

Even if they do need to interact across organizations and pay for API calls or data, the costs are unlikely to be paid autonomously by the agents. Usage-based costs are often abstracted by software vendors. This is how enterprise software stacks operate. Platform providers negotiate customized arrangements with data providers, compute vendors, and other infrastructure partners, bundling access rights into platform costs and passing them as a single aggregated line item.

Furthermore, they can achieve units of economic efficiency that no single agent can autonomously replicate. Computing resources are procured through reserved capacity agreements with AWS, Azure, or GCP. Model inference pricing is based on bulk agreements with companies like Anthropic, OpenAI, or Google. Data augmentation is handled via suppliers like Bombora or Clearbit. All these are pre-negotiated and abstracted.

In other words, 40,000 API calls, model inferences, and data queries do not generate 40,000 separate payments but produce a single invoice. The granularity of consumption is always different from the granularity of settlement, and enterprises tend to prefer maintaining this separation.

Consumer Agents Will Be Responsible for Coordination, Not Consumption

While commercial agents may not transact autonomously—since enterprises won’t allow it—consumer agents also won’t transact autonomously because people don’t want them to.

An example often cited by advocates of intelligent commerce: you ask your agent to book a trip to Tokyo. It searches hundreds of hotels, cross-checks reviews, checks your calendar, applies your preferences, and then automatically books the room. You don’t have to do anything. Of course, proponents of agent-based business models extend this user experience to nearly all consumption domains, from groceries to home goods to clothing.

The problem is, preferences are not static. They are reflected in the decision-making process itself. When booking a hotel, you’re not just looking for the lowest price. Your judgment reflects your mood, context, risk tolerance, and other qualitative factors you may not even be aware of before viewing options.

In practice, agents search, ask follow-up questions, and return options. You review pictures, inquire about the surroundings, maybe read some reviews. Then you make a choice and authorize the agent to use your stored credit card information for payment. In other words, the agent is merely a research assistant, not an independent economic actor.

Except for some predictable repeat purchases, this user experience is likely to be consistent across nearly all consumption domains, because consumer decisions rarely depend solely on price. The entire consumer economy is built on product differentiation. Whether it’s clothing, hotels, home goods, or groceries, decision-making involves countless qualitative factors that cannot be captured by intelligent agents—and more importantly, these factors exist within the discovery process itself.

During the discovery phase, agents will serve as powerful coordinators, but at critical moments, they will hand decision-making back to humans. Semantically, this is not agent commerce, nor does it require new payment channels.

The True Advantage of Crypto Payments: Bottom-Up Agents

While in the next five years, these two types of agents may account for over 95% of deployment, there is a third category worth noting.

In recent months, a new type of bottom-up agent has emerged. Driven by the OpenClaw phenomenon, these agents belong to a fundamentally different category. Unlike the aforementioned commercial and consumer agents, they are truly autonomous actors operating independently of any organizational entity. These agents need to make actual payments, and the granularity and frequency of payments make manual authorization impossible. Although the bottom-up agent economy is currently very small, it is likely to grow rapidly as some unforeseen emerging use cases appear.

Therefore, only in this very narrow context does the debate over whether crypto payments or card networks are the best underlying infrastructure make sense. While many cite technical advantages of crypto payments, I believe their ultimate reason for winning may be something else—permissionless.

Today, both payment methods are not optimized for agent commerce from a technical standpoint. Blockchain theoretically offers better unit economics for microtransactions, but it lacks identity verification and risk scoring mechanisms—features that could become especially important in the future of agent economies. Additionally, while real-time settlement is often mentioned, it simply means fraud transactions are settled immediately on-chain. Conversely, card networks have complex fraud detection systems and tokenized credentials that agents can inherit, but these tools are trained on human behavioral patterns and cannot be directly mapped to autonomous agent transactions. Moreover, for cross-border transactions, agents are also limited by the settlement times of card networks.

Perhaps counterintuitively, the reason crypto payments might become the default infrastructure for such agents is because blockchain is open, permissionless, and unregulated.

This is its ultimate structural advantage. While I believe existing card networks like Visa and Mastercard will continue to adapt through initiatives like Visa Intelligence Commerce and Mastercard’s AgentPay, they are ultimately listed companies bound by compliance, customer onboarding requirements, and partnerships with institutional counterparts. Blockchain, on the other hand, has no such restrictions. Anyone can develop on blockchain, and any agent can transact without approval.

Intuition suggests that emerging, experimental categories will develop where friction is minimal.

The Bottleneck Is Not Infrastructure, But Ourselves

However, the longer-term question is how this experimental pace can ultimately lead to greater impact. The bottom-up agent economy will only become truly popular when autonomous agent organizations significantly outperform human organizations enhanced by agents; this advantage must be substantial enough that top-down human restrictions on agents become a competitive disadvantage. At that point, agents will no longer be just automators of human tasks within closed environments but will become organizations themselves.

Yet, this future may still be far off. The bottleneck is not the technology itself. Moreover, what is truly “not suitable for machines” may not be the payment systems but everything else not designed for an autonomous agent economy: regulatory frameworks, bureaucratic institutions, legal structures, and social inertia surrounding human decision-making. These constraints have far-reaching implications beyond any technical detail in the payment stack. Unfortunately, protocol upgrades alone cannot solve these issues.

The scale of the agent economy will be enormous, with most transactions billed monthly.