.webp)
.jpg)
Coinbase CEO Brian Armstrong believes artificial intelligence could become one of the strongest catalysts for cryptocurrency adoption—not because AI agents will speculate on digital assets, but because autonomous software will need a practical way to hold, receive and spend money.
In a recent post on X, Armstrong rejected the idea that crypto companies must abandon blockchain technology to pursue opportunities in artificial intelligence. He instead presented AI and crypto as complementary technologies: AI supplies programmable intelligence, while blockchains and stablecoins provide programmable financial infrastructure.
Armstrong predicted that AI agents could eventually conduct more daily transactions than humans. These transactions may include purchasing data, accessing software, paying for computing resources, hiring specialized agents and executing financial tasks on behalf of individuals or businesses.
The prediction is ambitious, and the agent economy remains at an early stage. However, Coinbase has already turned the thesis into a product strategy involving AI-compatible wallets, stablecoins, its Base blockchain and the x402 internet-payment protocol.
Why AI agents need a new payment system
Most existing payment systems were designed around people and registered businesses. Opening a bank account generally requires a recognized legal identity, compliance checks and human authorization. Credit-card payments similarly depend on account credentials, billing relationships and interfaces created for human customers.
An autonomous agent cannot independently satisfy those requirements in the same way a person or incorporated business can. It can operate under authority delegated by its owner, but conventional payment infrastructure still introduces friction when software needs to make thousands of small, time-sensitive purchases.
Consider an AI research agent preparing an investment report. It might need to pay separately for market data, company filings, news archives, web searches and a specialized analytical model. A travel agent could purchase access to flight schedules, hotel inventory, maps and local transportation data. A software-development agent might need cloud computing, code repositories, security scans and model inference.
Traditional subscription and invoicing systems are inefficient for these machine-to-machine workflows. An autonomous program may need a single API request costing a fraction of a dollar rather than a monthly subscription. It may also need to complete that purchase instantly, without filling in a checkout form or waiting for an invoice to be approved.
Stablecoins and blockchain wallets offer an alternative. A wallet can be controlled programmatically, transactions can settle around the clock and smart-contract rules can limit how funds are used. Stablecoins are particularly relevant because their comparatively stable value makes them more suitable for pricing services than volatile assets such as Bitcoin.
Armstrong’s argument is therefore less about AI agents becoming crypto investors and more about crypto becoming the settlement infrastructure through which software pays for digital resources.
Coinbase’s x402 protocol connects payments to web requests
A central part of Coinbase’s strategy is x402, an open payment protocol introduced in 2025. The protocol revives the HTTP 402 “Payment Required” status code and converts it into a machine-readable payment process.
Under the x402 payment flow, an agent first requests a resource from a website or API. If payment is required, the server responds with payment instructions. The agent’s wallet creates an authorization, a facilitator verifies and settles the payment, and the server releases the requested resource.
The payment effectively becomes part of the normal internet request instead of a separate checkout process.
This architecture can support pay-per-request APIs, digital-content paywalls, computing services and other forms of usage-based billing. It can also remove the need for every agent to create an account, obtain an API key or maintain a separate subscription with every provider it uses.
Coinbase initially positioned USDC on Base as a natural combination for x402 because stablecoins provide predictable pricing while lower-cost blockchain networks make small transactions more practical. The protocol has since expanded across multiple networks and can support different digital assets.
Importantly, x402 is not supposed to remain a Coinbase-only system. Coinbase has described it as an open standard, and the project has been placed under an independent foundation within the Linux Foundation. Its long-term value will depend on whether competing wallet providers, payment facilitators, cloud platforms and service operators adopt compatible implementations.
Early commercial infrastructure is taking shape
Coinbase is building several products around what it calls the agentic economy.
Its Agentic Wallet infrastructure gives software agents the ability to receive, spend and trade digital assets within predefined limits. Developers can establish maximum transaction sizes and session-level spending caps, while private keys are separated from the language model itself. Coinbase says transaction monitoring and compliance controls can also be applied before payments are completed.
These controls are essential. Giving a language model unrestricted access to a wallet would expose users to prompt injection, software errors and malicious service providers. A workable agent-payment system therefore needs more than a wallet: it requires identity controls, permissions, transaction simulation, audit logs, service verification and reliable ways to suspend an agent.
Coinbase said in February that x402 had already processed more than 50 million transactions. Base subsequently reported that x402 recorded approximately 3.1 million transactions worth $1.2 million on its network during the 30 days ending May 29, 2026. Agents were reportedly purchasing model inference, browser sessions, search results, market data and research services. These figures demonstrate activity, although they do not necessarily prove broad commercial adoption. Base’s ecosystem report was published by the network itself and should be viewed in that context.
Coinbase has also worked with AWS to let publishers using CloudFront and AWS Web Application Firewall accept x402 payments from automated clients. The integration could allow publishers to charge agents for individual requests instead of either blocking automated traffic or allowing it to access content for free. Coinbase announced the AWS integration in June.
Stablecoins may benefit more directly than volatile tokens
If machine payments expand, stablecoins are likely to capture the clearest initial demand. An agent buying a five-cent data query needs a reliable unit of account. Paying with an asset whose value can change materially between pricing and settlement would create unnecessary accounting and risk-management problems.
Networks capable of processing large numbers of low-value transactions could also benefit. Base, Solana and other high-throughput networks may compete on transaction costs, speed, reliability, developer tools and stablecoin liquidity.
The investment implications for other crypto assets are less direct. Bitcoin may benefit from broader crypto adoption and increased institutional acceptance, but it is not necessarily the most efficient instrument for high-frequency agent payments. Ethereum and other smart-contract platforms may gain if agent transactions generate greater demand for settlement, identity, tokenization and decentralized applications.
Oracle networks could provide agents with verified external information, while decentralized computing networks might sell processing capacity to autonomous software. Nevertheless, the existence of an AI narrative does not automatically create sustainable demand for every AI-branded token.
The most durable value is more likely to accrue to infrastructure that solves measurable problems: stable settlement, secure wallets, inexpensive execution, reliable data, agent identity, policy enforcement and fraud prevention.
Transaction numbers require careful interpretation
High transaction counts can make agent-payment adoption appear larger than it is. Software can generate enormous numbers of extremely small transfers, and subsidized network fees make repetitive activity inexpensive.
A July 2026 population-scale study of x402 activity examined approximately 136.7 million Base settlements representing $44.1 million in value. The researchers found that payment activity was highly concentrated. They classified 21.2% of settlements as fictitious and another 63.78% as internal transfers within linked clusters.
The paper is a preprint and has not completed peer review, but its findings reinforce an important distinction: transaction volume does not necessarily equal independent economic demand.
For adoption to become convincing, analysts will need to examine the number of unrelated buyers and sellers, repeat customer behavior, revenue earned by independent services and whether transactions pay for useful outputs rather than promotional or internally generated activity.
Armstrong’s forecast that agents will out-transact humans could ultimately prove correct on a transaction-count basis. A single agent may generate hundreds of payments while completing one task. The more meaningful question is how much genuine economic value those transactions represent.
Security and accountability remain major barriers
Autonomous payments combine the unpredictability of AI models with the irreversibility of blockchain settlement. That combination introduces risks that conventional payment systems were not built to handle.
A compromised agent could overspend, send funds to a malicious address or purchase fraudulent services. Prompt-injection attacks could manipulate an agent into ignoring its intended instructions. A faulty facilitator could approve an invalid payment, expose merchant services without receiving funds or misuse the assets and transaction fees it manages.
Researchers studying 15 x402 facilitators reported authorization and implementation weaknesses affecting every system they tested. Their July 2026 security paper described potential attacks involving unpaid purchases, asset theft, service disruption and abuse of subsidized transaction fees. The researchers said affected providers were notified and that mitigations, including changes by Coinbase, were subsequently introduced.
Financial regulators are also paying attention. The Bank of England’s July 2026 Financial Stability Report identified authorization, traceability, fraud detection, liability, resilience and payment reversals as unresolved issues in agentic finance. It also highlighted a fundamental tension: AI systems can be probabilistic, while payment infrastructure must produce predictable and legally certain results.
For that reason, commercially viable financial agents will likely operate within tightly defined authority. Spending limits, approved-service lists, transaction simulations, human confirmation thresholds and emergency controls will be necessary even when routine low-value payments become autonomous.
Coinbase is betting on infrastructure, not just speculation
Armstrong’s broader message is that crypto should be viewed as general-purpose infrastructure rather than a market competing with artificial intelligence for capital and attention.
Coinbase is positioning itself across several parts of that infrastructure. It operates the Base network, provides wallet technology, supports USDC markets, offers payment-facilitation services and maintains relationships with both developers and businesses. If agentic commerce expands, Coinbase could benefit without needing any single consumer-facing AI agent to dominate the market.
The company’s strategy remains speculative. Traditional banks and payment processors could develop their own machine-compatible systems, while regulated tokenized deposits or instant-payment networks may compete with stablecoins. Enterprises may also prefer permissioned payment environments that provide stronger identity, dispute-resolution and compliance guarantees.
Even so, the underlying problem Armstrong identifies is real. Autonomous software needs a way to purchase digital services at machine speed. Conventional accounts, subscriptions and manual checkouts are poorly suited to millions of small, automated interactions.
Whether crypto becomes the dominant solution will depend on security, regulation, cost and genuine demand. But Coinbase is making a clear bet: if AI agents become independent economic actors, programmable money will become as important as programmable intelligence.
