The card network economics problem that might push AI agents to crypto

Interesting framing from CoinDesk on why AI agents might default to crypto rails:

An AI agent researching a topic might make 6 API calls in a single task — real-time news ($0.002), onchain data ($0.004), cross-referencing ($0.001), etc. Total cost: under $0.02.

Stripe's minimum fee is ~$0.30 per transaction. Running those 6 payments through cards would cost 100x the value of the payments themselves.

The catch is that agents also need to operate across chains. If an agent needs Solana data, Arbitrum execution, and Base settlement, it can't be manually bridging between each step. This is where cross-chain aggregators like SODAX become infrastructure rather than convenience — agents need unified execution layers that work programmatically.

Visa's Trusted Agent Protocol and Mastercard's recent Santander integration suggest TradFi sees this coming too. The split might end up being: regulated human commerce on cards, machine-to-machine on stablecoins.

What's your read — is the micropayment economics argument compelling, or are there other factors that matter more?