In today’s newsletter, First Digital’s Vincent Chok unveils the rise of “agent finance,” in which AI agents go beyond advice to execute financial transactions, making crypto the essential financial backend of this machine-driven economy.
Next, in “Ask an Expert,” we asked three leading AI systems – Grok, Gemini and Claude – two questions about AI payment use cases and the steps needed for scalability.
Note: Responses were generated by AI assistants and reflect each model’s perspective. It should not be construed as financial or legal advice.
-Sarah Morton
AI Agents in Crypto: What Advisors Need to Know
The explosive growth of AI agents
AI agents have become one of the hottest trending topics over the past year. A recent PwC survey of more than 300 companies found that 79% of them are already adopting AI agents in some form. This explosive growth reflects a broader shift: AI agents are moving from advisory to execution roles.
Initially deployed to facilitate chatbot services and co-piloting roles, AI systems now actively plan, decide and act on predefined parameters defined by humans, including financial transactions. The result is the early formation of a “finance agent”. This is a new primitive in which AI agents essentially execute financial actions according to predefined rules such as limits, permissions, and goals.
Breaking down agentic finance
Agentic finance can be understood at three levels. The agent commerce layer focuses on discovery and decision making. For example, an AI agent can search for the best hotel deal for an upcoming trip. The agent payment layer manages execution, where the agent completes a transaction once approved.
Finally, the asset management layer represents the full stack, where the agent can manage portfolios, manage payments, and dynamically optimize financial strategies based on real-time market trends. While this may seem like we are granting complete autonomy to AI agents, this is not the case. This is conditional delegation, in which users maintain control through constraints while offloading execution.
Theoretically, AI agents have a use case in the financial space; however, they do not integrate seamlessly with existing traditional financial infrastructure. Structurally, AI agents do not have direct access to global banking networks and are designed to operate 24/7. It is in this structural mismatch that cryptography comes into play.
Stablecoins provide AI agents with access to programmable, always-on money, blockchains enable instant and global settlement, and crypto wallets provide permissionless access to funds. Essentially, these components form a financial layer better suited to machine-driven activities. Crypto is thus increasingly becoming the infrastructure of autonomous systems, rather than just an asset class.
AI Agent Use Cases
The first implementations are already visible. Machine-to-machine payments powered by API access and data providers have made inter-merchant exchanges stronger and faster. In the consumer context, self-commerce has allowed users to optimize their retail searches, using agents to get the best deals on travel, subscriptions and purchases.
Meanwhile, in crypto-native environments, trading agents are widely deployed for portfolio management, return optimization and trading strategies. On the enterprise side, supply chain management and supplier payments were easily automated through AI agents, reducing errors and resource expenditure. At this stage, most activity remains business-to-business and infrastructure-focused, rather than consumer-focused.
Beyond use cases, AI agents also play a critical role in creating new investment categories as well as the demand for crypto itself. As AI agents cannot run on existing infrastructure rails, demand is increasing for agent-native wallets, stablecoin payment rails, and data or compute marketplaces.
Coinbase, for example, launched x402, an open payments protocol designed for agent-native transactions. This shift is particularly relevant for micropayments, where high transaction volumes and low value make traditional rails inefficient. For the first time, non-human users are participating in the financial system and driving activity. AI agents have become a new class of “users” for crypto networks.
Risks and future prospects
Despite this momentum, we are still in the early stages, and there are risks and limitations. Security is the primary concern, especially when dealing with malicious or exploited agents executing unintended transactions. Issues regarding authorization, liability and regulatory treatment are still under review and are actively being defined. For widespread adoption, we need to build user trust. This requires regulatory clarity from all stakeholders involved, so that projects can develop with clarity and confidence while protecting funds and user interests.
Over the next twelve months, this technology will continue to grow and mature. Important signals include growth in agent-driven transaction volume, the emergence of agent-native wallets and payment protocols, and deeper integration between stablecoins and AI-driven systems. Finally, regulatory clarity will strongly determine the pace and scope of adoption across different sectors and areas.
In conclusion
AI agents are not a theoretical concept; they already execute transactions in limited environments. As the trend grows, crypto is increasingly emerging as the financial backbone of machine-driven economies. For now, this is a long-term infrastructure and thematic piece; however, this is changing as adoption rates increase. Advisors should view it as a driver of the next wave of crypto utilities.
– Vincent Chok, CEO and co-founder, First Digital
Ask an expert
This week we’re doing something a little different. Instead of one expert, we have a panel of experts: AI experts. Below, we asked three leading AI models two questions about the present and future of AI payments. While there are common themes, particularly around what is needed for future growth, there are also clear differences. We hope you find this experience as fun and challenging as we do.
Q1: What AI payment use cases are you seeing today?
Grok xAI |
Fraud detection is the most mature use case: AI detects anomalies missed by rules-based systems. Intelligent payment routing dynamically selects the best train based on cost and speed. Emerging agentic payments enable autonomous AI to handle B2B cash tasks and machine-to-machine micropayments using stablecoins, with Visa, PayPal and Mastercard starting to verify these agents. |
|---|---|
| Gemini Google | Three advances in “agent trading” stand out: autonomous DeFi cash and yield management where AI executes cross-chain trades in real time; the “inference economy” in which agents instantly settle micropayments from machine to machine via stablecoins; and hyper-personalized wealth implementation – AI-driven direct indexing managing split trades and daily tax-loss harvesting at a scale that no human advisor could manage manually. |
| Claude Anthropic | AI payments are moving from concept to infrastructure faster than most advisors realize. Clearest use cases: Programmatic billing for AI services and agents that autonomously spend funds on computation or subtasks without human approval per transaction. On the wealth side, AI is increasing rebalancing and tax-loss harvesting, with humans still privy to most material decisions today. |
Q2: What is needed for AI payments to evolve?
Grok xAI |
Three gaps need to be filled: a standardized agent identity with proof of cryptographic authorization and clear accountability rules; governance and explainability for regulators, including AML compliance in automated wealth flows; and high-quality real-time data infrastructure connecting fiat and crypto currencies. Advisors must demonstrate that AI augments rather than replaces fiduciary duty. |
|---|---|
| Gemini Google | Three elements must converge: verifiable agent identity using DIDs and cryptographically signed mandates so institutions can distinguish authorized AI from malicious bots; standardized governance protocols with universal spending limits and kill switches applied at the protocol level; and legal liability frameworks clarifying who is responsible when an AI executes a faulty transaction or falls victim to a rapid injection attack. |
| Claude Anthropic | Three things are truly missing: identity, accountability and spending control. AI agents don’t have a standardized identity layer – that’s the fiduciary risk behind all the excitement. Programmable guardrails natively integrated into the payment layer (like a corporate card with strict limits and audit trails) are needed, as well as regulatory clarity on stablecoins. The technology is largely ready. What’s lagging behind is governance – and that’s an opportunity for advisors who get ahead of the curve. |
