Key Highlights
- Coinbase has outlined its approach to Agentic Finance, or AiFi.
- AI agents can interact with Coinbase accounts within permissions and limits set by users.
- Coinbase’s x402 payment standard allows software to pay for APIs, data, and other digital services.
Coinbase is developing tools that allow artificial intelligence agents to interact with crypto accounts, execute transactions, and make payments with limited human involvement.
The company made an announcement on August 27, describing the broader concept as “Agentic Finance,” or AiFi.
The products cover two main areas: AI-assisted management of financial accounts and automated payments between software systems. In both cases, Coinbase says users can establish permissions and limits before an agent carries out a transaction.
AI agents can interact with crypto accounts
One part of the strategy involves connecting AI applications to financial accounts.
Through Coinbase for Agents, users can connect applications such as ChatGPT, Claude, and Cursor to Coinbase accounts using the Model Context Protocol (MCP).
Users can determine what an agent is allowed to access and which actions it can perform. Coinbase also allows users to create a separate account for an AI agent rather than connecting the software to their primary portfolio.
One potential use case is an agent monitoring a portfolio and executing a transaction when a predefined condition is met.
The setup does not give the software unrestricted control. The account holder determines the permissions and limits under which the agent operates.
Coinbase Advisor focuses on portfolio management
Coinbase is also applying AI to portfolio analysis through Coinbase Advisor.
The company describes Advisor as an SEC-registered investment adviser available to eligible Coinbase One users in the United States, with the rollout occurring gradually. The service provides portfolio analysis and investment guidance through a conversational interface.
Coinbase said its research found that 70% of crypto investors would allow AI to help manage a portfolio if appropriate safeguards and a trusted brand were available.
That figure comes from Coinbase’s own research and is not an independent measure of investor sentiment across the broader market.
Advisor also represents a different use of AI from autonomous trading, as the product focuses on analysis and guidance rather than unrestricted control over customer assets.
x402 brings automated payments into the model
Coinbase’s other major AI-related initiative is x402, a payment standard designed for transactions between software systems.
The protocol allows an AI agent or other software to make a payment when it needs access to a service. Potential applications include paying for an API request, accessing data, purchasing research, or obtaining computing resources.
Coinbase says x402 has processed more than 205 million transactions involving approximately 200,000 sellers.
The company also says Coinbase has facilitated about 67% of the activity.
Those figures come from Coinbase and describe activity within the x402 ecosystem rather than conventional consumer payment volume.
A crowded field, not a Coinbase exclusive
Coinbase isn’t the only company building this kind of infrastructure, and its own numbers come with a caveat worth noting: The Crypto Times previously reported that x402’s transaction growth has leaned heavily on high-frequency, low-value payments averaging around $0.52 each, small enough that requiring manual approval on every transaction would often cost more in a person’s time than the payment itself is worth.
That gap is part of why rivals are racing to build their own authorization layers rather than simply adopting Coinbase’s approach. Google has donated its AP2 “mandate” system, which defines what an AI agent is allowed to spend and under what limits, to the FIDO Alliance, while Visa and Mastercard have each signaled support for open agentic-payment standards through their own initiatives.
AWS, meanwhile, has built USDC-based agent payments directly into Bedrock AgentCore in partnership with both Coinbase and Stripe, meaning Coinbase’s rails are being used alongside, not instead of, competing infrastructure.
Businesses can accept payments from AI agents
Coinbase is also developing payment infrastructure for businesses through Coinbase Business.
Eligible businesses can accept USDC payments from AI agents, allowing software to pay for digital services without requiring a person to complete a conventional checkout process.
Traditional online payment systems generally assume that a person enters payment details, completes authentication, or approves a transaction.
Agent-based systems instead allow software to make payments when it has the necessary authorization and the transaction meets predefined conditions.
The model could therefore allow stablecoins to be used for machine-to-machine payments involving digital services.
USDC is central to the payment infrastructure
Stablecoins form an important part of Coinbase’s approach because they can be transferred programmatically across blockchain networks.
Coinbase says USDC accounts for 99% of onchain agentic commerce, while x402 represents 97% of agentic protocol usage.
Both figures are Coinbase estimates and should not be treated as independent measurements of the overall AI-payment market.
Coinbase is also positioning Base as an environment for these transactions, citing its transaction costs and existing developer ecosystem.
The company’s AI initiatives therefore rely largely on infrastructure already used for crypto trading and stablecoin transfers.
Automated finance brings new risks
Allowing AI agents to move funds or execute trades introduces risks that do not arise when every transaction requires direct human approval. These include unauthorized access, incorrect instructions, security breaches and decisions based on incomplete information.
There are also questions about who would be responsible if an AI agent makes an erroneous transaction that causes a financial loss.
Coinbase’s use of separate accounts, spending limits and user-defined permissions can limit an agent’s access, but these measures do not eliminate the risk of errors or unauthorized activity. Regulators will also need to determine how existing financial rules apply when software interacts directly with trading and payment systems.
Coinbase’s approach combines AI agents, crypto accounts, stablecoin payments, and blockchain settlement, in a race where Google, Visa, Mastercard, and AWS are all building comparable pieces of the same puzzle.
Whether Coinbase’s version gains a lasting edge will depend less on its own announcement and more on how these competing standards, and the regulatory questions around them, get resolved.
Also Read: UK to Give Bank of England New Objective on Stablecoins and Digital Payments
