Sandy Kaul, Franklin Templeton's Head of Digital Assets and Innovation, published a piece this week called "Agentic AI—The Killer Use Case for Blockchain and Crypto." Her argument, stripped of the TradFi packaging: card networks charge 2%–3% plus a flat fee of about $0.30 per transaction. An AI agent buying a second of compute or a single data query costs roughly $0.001. Run that through a card rail and the fee is bigger than the purchase. Something else has to settle it.
That's not a new observation to anyone who's shipped a payment skill for an onchain agent. It's the exact reasoning behind [x402](https://satohub.ai/directory?q=x402), the pay-per-request HTTP standard Coinbase open-sourced and handed to the Linux Foundation, and it's why Stripe and Visa are building their own machine-payments protocol in parallel. What's notable isn't the math — it's who's saying it. A $1.5 trillion asset manager just told its audience that agentic commerce is a plumbing problem, and the plumbing that clears is the one that can price a fraction of a cent.
The fee math, again
Kaul's framing: agent-to-agent commerce needs "software paying software," at volumes and denominations no card network was built for. She points to two numbers that do the actual work in the piece — agentic commerce sized at $3–5 trillion by 2030, and a stat that 38% of organizations expect AI agents to be functioning as team members by 2028. Both are projections, not receipts, and Franklin Templeton attaches the standard risk disclosures — these are estimates, concepts may not materialize, crypto carries total-loss risk. Read the thesis, not the forecast.
Where the piece is on firmer ground is the plumbing itself. Card infrastructure is built around authorization holds, chargebacks, and settlement windows measured in days — overhead that makes sense for a $40 grocery run and none at all for ten thousand micro-charges an hour. A blockchain settlement layer with a native fee-bearing asset doesn't have that overhead. That's a description of what x402 and similar rails already do, not a prediction.
What she's actually pointing at
Kaul cites x402 by name and nods to Solana's SOL as an example of a network token that exists to price transactions, not to be a stock proxy. That's the tell that this is a plumbing argument dressed as an investment thesis: the case isn't "buy the token," it's "the network that settles agent payments needs a fee asset, and that's a structurally different thing from equity." Builders don't need the investment framing to find the observation useful — an agent that spends money needs a wallet, a payment rail priced for its transaction size, and a settlement layer that clears in seconds, not days. That's three concrete requirements, not a chart pattern.
The part TradFi coverage skips
What the piece doesn't do — and what a builder actually needs — is name which rails are live versus which are roadmap. x402 is shipped and in production use today. Stripe and Visa's Machine Payments Protocol is announced, not deployed. "Software can pay software" is true for at least one of those and aspirational for the other. Distinguishing claimed from shipped is the whole game here, and it's the same discipline [Sato Hub's builder library](https://satohub.ai/build) applies to every payment rail it tracks — what's live, what's self-reported, what's actually been checked.
What to watch
Watch whether more asset managers follow Franklin Templeton in framing agentic AI as an infrastructure thesis rather than a model-capability story — that's a signal the payment-rail argument is landing outside crypto-native circles. And watch the actual deployment gap between x402-style rails already clearing agent payments and Machine Payments Protocol's still-unshipped promise. The fee math is the same either way; only one of them is live.
Sources
- ▸[Franklin Templeton Says Altcoins Are Key to the Agentic AI Trade — The Defiant](https://thedefiant.io/converge/tradfi-and-fintech/franklin-templeton-says-altcoins-are-key-to-the-agentic-ai-trade)