Imagine giving an AI assistant $10 to gather data for an investment decision. Along the way, it can buy access to information. But whether to also let it execute the trade itself is a separate question.
"Allowing an agent to spend 40 cents and entrusting it with portfolio management are completely different things," Zoomex emphasizes.
The question of responsibility runs through the entire "State of AI Agent Payments 2026" report. From July 23 to August 26, researchers tracked 6.4 million payments via the x402 protocol on the Base and Solana networks.
In total, $119,947 passed through them.
The average payment was about 1.9 cents. More than 90% of transfers were under one cent.
What AI agents actually buy
A separate analysis of the x402 catalog showed where these small purchases go. Among the most in-demand resources are web search and checking available premium seats with airlines.
At the Zoomex trading platform, they expect the number of transactions to grow faster than the average payment size over the next year. Buying information gives agents a useful task, while each purchase risks a relatively small amount of money.
Moving to investment decisions requires determining how much freedom to give the program.
Of the 19 products and scenarios reviewed, 14 could operate within pre-agreed permissions. That is, some operations took place without confirmation each time.
The scope of such authority varied across products.
Zoomex's own AI integration offers a different level of delegation. Through the MCP service, which connects AI applications with the exchange's tools, assistants like Claude or Cursor can study the market and prepare orders.
"Our MCP allows AI assistants to obtain market data, analyze positions, and prepare trading instructions, but the user retains control — the trade is executed only after their confirmation," Fernando Lillo Aranda told the researchers.
Separately, Zoomex recorded throughput of up to 60,000 order requests per second for approved institutional accounts. This figure describes permitted programmatic traffic. It does not account for actual trades or decisions made by AI.
In the readiness assessment from the report, micropayments scored 60.7 out of 100, while trading and investment agents scored 39.1.
"Over the next twelve months, we expect growth in adoption in clearly defined tasks: rebalancing within agreed limits, collateral management, or order execution under specified conditions. Fully autonomous investment decisions will come later," Zoomex noted.
The progress metric proposed by the company puts the account owner at the center:
"The key metric is not only how much money agents move. It is how much responsibility people are willing to delegate to them."
Cryptalist analysis: The agent payments market is still in its nascent phase: a turnover of $120,000 with an average ticket of less than two cents indicates a predominance of microtransactions for data rather than real capital management. The gap in readiness scores (60.7 versus 39.1) reflects a fundamental limitation — payment infrastructure is outpacing trust in autonomous investment decisions. For institutional participants, the key growth driver will not be expanding AI authority, but rather the development of controlled delegation mechanisms with limits and mandatory trade confirmation.