Agent payment protocols have multiplied, but real transaction volume remains close to zero because agents already use API keys to access paid services.
2
Crypto fits agent payments because it supports small transactions without traditional payment fees or chargebacks, although wallets are still difficult to set up.
3
Agents need better models, trusted service directories, built-in payment tools, and pay-per-use billing before autonomous purchasing can become practical.
Summary
Jan Curn examines why the growing number of agent payment protocols has produced almost no real volume. Apify, which runs a marketplace of about 80,000 tools and paid creators almost $2 million in the previous month, sees customers using its services through API keys rather than autonomous discovery and payment. Curn breaks the problem into four stages: an agent must need a service, be given a way to pay, discover a service it can trust, and complete a purchase. API keys remain easier than autonomous payments, crypto wallets are unfamiliar, and random services lack trustworthy reputation signals. Fixed-price payments also do not fit jobs with variable usage. Curn describes how Apify tested newer x402 payment schemes, including batch settlement, to support pay-per-use billing. He expects always-on agents, better models, less generous token subsidies, trusted directories, and payment integration inside major agent platforms to increase adoption.
Agent payments are competing with API keys, not credit cards
Most agents already access paid services through accounts created in advance, API keys, MCP connections, and a user's credit card. Curn says this is the real alternative to autonomous payment because it is straightforward and gives the agent a fixed set of tools. For autonomous purchasing to happen, people must want agents to discover services as tasks unfold. Better models may increase trust gradually, with budgets growing from small amounts to larger ones. Always-on personal agents could also receive their own budgets. Curn expects token subsidies to decline, which may make buying an existing service more economical than building every capability from scratch.
Crypto fits small agent transactions better than traditional payment methods
Curn argues that crypto is well placed for agent payments because it has no single controlling company and can make small transactions more economically. Credit cards, PayPal, micropayments, and bank debits can be too expensive for small agent purchases. Traditional payments also allow chargebacks, which Curn says would be difficult for online services facing an unlimited number of agents. Crypto does not have chargebacks, although merchants can choose to issue refunds. He points out that the two payment systems he describes as most popular, x402 and MPP, use stablecoin-based crypto payments.
Real agent payment volume is tiny despite many available services
Curn shows transaction data from x402 and MPP monitoring sites. In the previous 30 days, x402 had about $1 million in payment volume, while MPP had about $1,500. He says this is effectively non-existent compared with the attention around the protocols. The problem is not necessarily a lack of services. Apify, Exa, Perplexity, and Farscraper already make far more than that per month from agents using their tools, but those purchases happen through API keys rather than autonomous discovery and payment.
Wallet setup remains a barrier for ordinary users and agents
After an agent has a reason to buy a service, someone still has to set up the payment system. Curn says crypto remains opaque to many people because of its terminology and wallet setup. Using Coinbase for trading does not mean someone has given an agent its own wallet. He expects better tooling and user experience to reduce this friction. A larger change would come if major agent platforms such as Claude or ChatGPT connected crypto payments directly to an existing account balance, allowing a user to give an agent a budget without handling the wallet details themselves.
Autonomous agents need trusted ways to find services
An agent buying from a random URL needs evidence that the service will deliver what it promises at the stated price. Curn compares this with how people build trust through recommendations, marketplaces, reviews, consumer protection agencies, and other intermediaries. The agent ecosystem lacks enough of these signals. Trusted directories and marketplaces can fill the gap. He cites Claude's reviewed connector directory, Perplexity's apps, and Apify's marketplace as examples where a middleman helps an agent decide which services are safer to use.
Early agent payment systems mainly supported a fixed price for one API call. Curn says that works for something like a weather forecast, but real jobs may launch compute or another agentic process whose cost varies with use. That requires pay-per-use billing. x402 added batch settlement, while MPP added sessions, to support this pattern. These payment schemes let an agent use a service and settle the resulting variable cost instead of paying only for a predetermined call.
Batch settlement solved a problem that earlier variable payments left open
Curn describes the payment flow as a request, a 402 response containing payment requirements, a signed retry, verification by a facilitator such as Coinbase, and later blockchain settlement. The earlier up-to payment scheme allowed variable costs, but the buyer could potentially spend the same money twice before settlement. Apify worked around this by charging an amount up front and refunding what the job did not use, which required trust in the seller and extra blockchain transactions. The newer batch settlement scheme uses cryptographic escrow on the blockchain so later microtransactions can be charged against reserved funds.
