# Why AI Agents Still Won't Pay for Anything _Apify's Jan Curn says agents can pay but don't: $1M on X402 and $1,500 on MPP in 30 days while API keys win._ **Published:** 2026-10-18 **Source:** https://www.startuphub.ai/ai-news/artificial-intelligence/2026/why-ai-agents-still-won-t-pay-for-anything --- Jan Curn told [AI Engineer](https://www.youtube.com/watch?v=-AOxpwVeIWw) that agentic payments, the ability for an AI agent to discover and pay for a service on its own without a pre-shared API key, are technically ready and practically unused. Curn, who runs [Apify](/startups/anything) and its marketplace of about 80,000 tools called actors, said the company paid almost $2 million to creators last month and has chased every new payment rail to let agents buy those tools directly. The bet made sense when he was the lone voice on agent payments at AI Engineer World Fair in San Francisco last year, and it looks obvious now that this year the same event had a dedicated track. The timeline explains why it feels crowded. Lightning Labs shipped L402 in mid-2020, Mastercard followed with Agent Pay in April last year, Coinbase shipped X402 in May 2025 which Curn called the most popular standard today, [Skyfire](/startups/skyfire) shipped KYA Pay with Visa in June, then Google shipped AP2, [OpenAI](https://www.startuphub.ai/startups/openai) and Stripe shipped ACP in September, Visa shipped TAP in October, Google and Shopify shipped UCP in January, and this year brought ACTP from Alipay, MPP from Stripe and Tempo, AMM from Alipay in April, APOP from UnionPay and an Agent Payment Protocol from OKX, plus another Agent Pay for Machines from Mastercard. None of it is moving money. Curn pointed to public scanners showing about $1 million in volume on X402 in the last 30 days and about $1,500 on MPP in the same window, both effectively zero. The supply is not the problem, he said, because directories already list [Apify](/startups/anything), Exa, Perplexity and [Firecrawl](https://www.startuphub.ai/startups/firecrawl) and agents do pay those vendors millions per month, but through fixed accounts and API keys. Agentic payments are not competing with Visa, they are competing with the API key, and the key still wins on convenience. That gap is starting to force a rethink beyond the protocol itself, with banks now being pushed to verify not just a payment credential but the agent, its permissions and the buyer’s original intent, according to PYMNTS. Curn broke the adoption problem into a funnel and walked it step by step. First is need. Most teams give an agent a fixed set of keys today and pay by card, so there is no reason to let it shop. That changes as models get better, agents run longer tasks, and always-on personal agents from Meta, OpenAI and others become normal, he said, and as token subsidies fade. He cited $100 per month for [Claude](https://www.startuphub.ai/ai-news/technology/2026/claude-code-s-latest-features) buying what he estimated as $14,000 worth of tokens if maxed out, a discount that removes any incentive to buy instead of build. Second is setup. Crypto is cheaper for micro-transactions and has no chargebacks, which Curn prefers for agent scale, but wallets and lingo remain opaque for most buyers. He said the big unlock would be if major agent platforms let users allocate credits directly, like telling Claude it can spend $50 on tools. Third is discovery and trust. You would not buy from a random URL claiming perfect weather data without a signal that it delivers, he said, and historically that signal came from knowing the seller, a referral, or a marketplace. Claude’s directory of connectors and marketplaces like Apify now play that middleman role. Last is the buy itself. The earliest schemes only handled a fixed price per API call, which covers a weather lookup but not a variable compute job that uses an LLM. Curn said X402 added batch settlement and MPP added sessions early this year to handle pay-per-use, and the path shows how rough it has been for early sellers. X402’s exact scheme in May required a 402 response, a signed payment header, verification via a facilitator like Coinbase, on-chain settlement and then work, which forces sellers to decide whether to do work before settlement and risk loss. An up-to scheme meant for variable pricing failed to prevent double spending because settlement happened later, so Apify worked around it with exact plus refund, which required agents to trust the seller and added extra chain fees. Batch settlement, which Curn said Apify added last week, escrows funds with crypto magic and then meters micro-charges, and that finally fits real jobs. The funnel is not theory anymore. Curn’s judgment is that the pieces for need, setup, trust and variable billing are lining up now, which means volume should come after distribution does, not after another protocol launches. Watch whether the platforms that own the agent also own the wallet. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.