Franklin Templeton's digital assets lead argues autonomous AI programs need near-instant, near-zero-cost settlement that card networks cannot deliver, pointing to fast blockchains as the practical rails
Autonomous AI software may become a major force behind blockchain usage, according to Sandy Kaul, head of digital assets and innovation at Franklin Templeton. In a lengthy July 22 post on X, Kaul argued that programs able to act, decide, and pay without constant human approval create a form of commerce that traditional card networks were never built to handle. For U.S. investors and developers watching both AI and crypto, the claim reframes how exposure to the next phase of AI might work.
Most portfolios seeking AI exposure still lean on shares of companies that build models and tools. Kaul's case is that this approach fits today's product cycle better than a future in which software agents negotiate and settle value with one another. That shift, if it scales, would put pressure on payment infrastructure rather than only on software valuations.
Card Rails Limits
Standard Visa settlement often takes one to three business days. Fees that are tolerable for human retail purchases become uneconomic when agents need to move tiny amounts many times per minute. Agentic commerce, as described in industry research, depends on settlement measured in seconds and costs close to zero. Those two requirements effectively rule out legacy card rails for high-frequency machine-to-machine payments.
Kaul pointed to Solana, Aptos, and BNB Chain as networks that already settle in seconds and therefore hold a structural edge when both sides of a transfer are algorithms. The argument is infrastructure-first: speed and fee levels matter more than branding when the counterparties never sleep and never wait for a bank batch window.
Industry Research Aligns
Kaul is not alone. On July 16, Visa's crypto unit and research platform Artemis released a joint report reaching a similar structural conclusion. Near-zero fees and rapid finality were framed as baseline conditions for agentic micropayments at commercial scale. The same report found that card systems designed for low-frequency human spending cannot meet those conditions.
Both Visa's crypto division and Stripe-backed Tempo introduced AI-oriented payment tools in March. Visa's offering aims to let agents complete same-day payments, narrowing part of the settlement gap its own research highlighted. Separately, Coinbase's x402 payment protocol recorded about $15 million in adjusted volume across more than 109 million adjusted transactions from its May 2025 launch through the date of the Visa-Artemis report. Those figures are an early signal of how often automated agents may already be moving value, not proof of a mature market.
Networks And Risks
Kaul did not name a single winning protocol. The focus stayed on properties already live on several chains: fast confirmation and low unit cost. For U.S. users and builders, that still leaves open questions about custody, smart-contract risk, stablecoin rails, geographic product availability, and how agents will authenticate counterparties. Choosing where value sits also raises familiar platform concerns; readers weighing where to hold or move funds can review practical steps on checking exchange safety before parking large balances alongside any agent-payment experiments.
Measurable context remains limited but concrete: multi-day card settlement versus second-level blockchain finality on the named networks, Visa-Artemis emphasis on near-zero fees, March product launches from Visa crypto and Tempo, and x402's reported $15 million adjusted volume with over 109 million adjusted transactions in the window ending around mid-July 2025. None of those data points alone proves mass adoption; they outline the performance gap Kaul and the joint report both describe.
Machine-to-machine payments change the economics of settlement because frequency and ticket size invert the human retail pattern. A person might buy once; an agent might pay for API calls, compute, data, or routing hundreds of times in a short span. When each transfer must clear quickly and cheaply, batch-oriented card systems add latency and fixed costs that compound. Blockchains that finalize in seconds reduce that friction, but they introduce different trade-offs: network congestion, fee spikes under load, bridge and wallet security, stablecoin redemption rules, and the need for reliable identity or reputation layers between agents. For developers and institutions in the United States, the practical path is less about declaring a single chain the winner and more about matching settlement speed, cost, custody model, and compliance constraints to the actual payment pattern agents will run.
