Agentic artificial intelligence (AI) is rapidly becoming one of the most significant paradigms in the new era of foundation models. By agentic, we refer to AI systems that can not only produce outputs but also execute actions in a given environment. Consider automation scenarios like responding to emails, interacting with a database, calling an API, or performing physical tasks in an embodied AI setting such as robotic tasks.

Unlike previous automation paradigms, agentic AI has the potential to enable agents to achieve semi- or full autonomy within their environments. We can envision a near future where tens of millions of semi-autonomous agents become integral to business and personal ecosystems. Crypto assets could serve as a fundamental representation of the world these AI agents interact with.

In other words, crypto holds the potential to become the currency of AI. But, while this notion is frequently promoted as a way to show the promise of crypto in powering a new intelligent economy, significant practical challenges lie ahead.

To understand the potential of AI agents using crypto, it’s essential to assess the fragmented state of the current agent market. The promise of semi-autonomous agents has triggered an explosion of frameworks for building them, far surpassing market readiness. While most organizations recognize the potential of autonomous agents, they remain in the early stages of adoption. As a result, we see numerous agent demos but very few production-ready products.

This is significant because the intersection of crypto and autonomous agents relies heavily on the maturity of the latter market. In a world where millions of autonomous agents augment our business and personal tasks, crypto could enhance these capabilities. There are several hypotheses about this augmentation, ranging from coordination to monitoring. But the greatest opportunity lies in the financial domain.

If we think of autonomous agents as extensions of our business and professional workflows, then e-commerce will be a crucial component of interactions between humans and agents, as well as between agents themselves. And crypto represents the most promising opportunity to unlock this potential.

Payments have been central to the evolution of the internet. PayPal unlocked many early e-commerce workflows, and platforms like Stripe have fully catalyzed this transformation. If AI agents become the fabric of the intelligent internet, financial transactions will be a key component of this new layer.

The programmability of crypto makes it an excellent fit for enabling payment capabilities in AI agents. But what scenarios make crypto ideal for agentic workflows while highlighting the limitations of fiat?

Imagine a situation where an agent engages another agent to complete a specific task such as paying invoices and following up with clients via emails. In this scenario, agents might receive numerous of those tasks, with payments unlocked as these tasks are completed( ex: an invoice is successfully paid).

The nascent state of the AI agent market is a challenge to making any of these scenarios work in practice. Therefore, the evolution of payment transactions in agentic workflows will likely not follow a straightforward path to crypto adoption and may undergo several iterations.

One often-overlooked obstacle to the adoption of crypto in multi-agent architectures is the cultural mismatch between the AI and crypto communities. Most teams driving innovation in agentic platforms have little to no connection to the crypto market, are not enthusiastic about Web3, and perceive crypto as a friction point for platform adoption beyond the crypto industry. This cultural disconnect extends to investors as well.

Many leading AI agent platforms were funded during one of the most challenging bear markets in the crypto space, a period when venture capitalists were particularly skeptical of crypto’s value. Additionally, organizations adopting AI agent platforms may still view crypto as a barrier to adoption within their operations.

While high-quality AI teams are emerging in the Web3 and AI agent space, this cultural friction should not be underestimated. Overcoming these challenges will require unlocking use cases that are impossible without crypto payments.

In the crypto world, the idea of AI agents conducting financial transactions is often discussed as an inevitability. While the promise is clear, the required infrastructure is lacking. Companies like Stripe built their businesses on decades of e-commerce innovation, supported by extensive merchant networks. This infrastructure does not yet exist for AI agents. As a result, early use cases for agents transacting may resemble the internet of the 1990s and will need significant evolution.

Let’s explore ideas for AI agent transactions using a hypothetical but plausible scenario. Imagine a group of agents across different geographies assigned 1,000 tasks, with payments unlocked upon completion of each one. To complete these tasks, some agents may need to collaborate with others. While some agents complete their tasks quickly and efficiently, others fail to deliver on time or effectively, leading to varying levels of conflict.

We have three options in today’s market:

1. Programmable Fiat

Programmable fiat platforms like Stripe offer a low barrier to entry for enabling financial transactions in multi-agent systems. Stripe’s recent announcement of an agent toolkit, integrated with many top frameworks, underscores this potential. The wide distribution of programmable fiat platforms, their large developer communities, and seamless integration with banking rails make them an attractive starting point for agentic payments.

However, programmable fiat presents limitations, such as challenges with cross-border payments, lengthy dispute resolution processes, and the complexity of performance-based payment triggers, making it suboptimal for agentic workflows.

2. Stablecoins

Stablecoins represent a natural intersection between AI and Web3. They address some of fiat’s limitations in agentic workflows, offering on-chain settlements, global availability, and enhanced programmability. In this scenario, agents could receive payments in stablecoins like USDC or USDT upon task completion.

However, stablecoins fall short in key areas. They are primarily payment tokens and lack functionality linked to agent performance. For example, in cases where agents fail to complete tasks, disputes must be resolved outside the stablecoin payment system.

3. A Native Agent Cryptocurrency and Payment Protocol

The true intersection of Web3 and agentic AI for financial transactions could emerge through new tokens and protocols tailored for this use case. These could extend stablecoin capabilities by integrating agent-specific functionalities.

In our scenario, payments could be made using a specialized asset that agents can stake for quality control. Slashing policies could penalize poor performance, while validators could resolve disputes based on task quality. Additionally, agents’ reputations could be directly tied to their token stakes. While this approach requires advancements in both Web3 and agentic AI, it is not as distant as it may seem.

Agentic AI has the potential to become one of the most fertile intersections between Web3 and AI. Semi-autonomous agents will become ubiquitous, requiring robust financial infrastructure for transactions. While programmable fiat may address initial needs, its limitations highlight crypto’s potential to shine.

Stablecoins offer a promising start, but they may not suffice in the long run. The rise of agentic AI is likely to drive the creation of custom tokens, wallets, protocols, and possibly blockchains optimized for agentic workflows. Agents could become the gateway for crypto to emerge as the true currency of AI.

Note: The views expressed in this column are those of the author and do not necessarily reflect those of CoinDesk, Inc. or its owners and affiliates.