Raoul Pal on why AI agents will need blockchain rails and which Layer 1s he thinks win

The core argument is not about crypto price action. It is structural. AI agents transacting with other AI systems need payment rails. Tiny, instant, high-volume machine-to-machine settlements. The legacy banking system was not built for that. Minimum transaction sizes, settlement delays, compliance friction, counterparty requirements. None of that is compatible with an AI agent paying another AI agent fractions of a cent thousands of times per second.

Pal's view is that Layer 1 blockchains, specifically Ethereum, Solana and Sui, are actually positioned for this because they are designed around programmable, permissionless, low-latency settlement. Not because of any crypto ideology but because the technical requirements of machine-to-machine economic activity happen to match what those networks already do.

The part I find most compelling is the framing around stablecoins as the denomination layer. AI agents are not going to transact in volatile assets. They need stable value transfer. The stablecoin infrastructure being built on these networks is the piece that makes the thesis practical rather than theoretical.

The honest caveat Pal himself acknowledges: the timeline is genuinely uncertain. AI agents transacting at that scale is probably not a 2025 phenomenon. Whether it is a 2027 phenomenon or a 2032 phenomenon changes the investment calculus significantly.

For people building in this space: are you designing for AI agent use cases yet or is that still firmly in the roadmap category rather than the current product category?