Rishin Sharma, the Solana Foundation's head of AI initiatives, outlined a plan for AI agents to get funding through a card-to-stablecoin route and pay for model responses one request at a time.
He presented the roadmap at Breakpoint 2026 in London on Nov. 17. The conference ran from Nov. 15-17 at Olympia London. The plan combines financial benchmarks for AI models with paid inference, the computing process that produces a model's answer.
Under the proposed billing approach, an agent would pay each time it calls a model instead of relying on a recurring monthly fee. The card-to-stablecoin route is intended to provide funds before the agent uses a paid service. Inference is the step in which the model generates a response after receiving a request.
Sharma discussed x402, a standard for payments between machines, and Solana payment channels. He said the channels could process as many as one million payments each second, describing a capacity rather than a tally of completed payments. He also described Pay.sh as a service offering paid access to more than 100 APIs.
His presentation named Google, Alibaba and ByteDance in connection with AI models available under request-based charges. The references concerned model access; the card-to-stablecoin route addressed the separate question of how an agent would receive funding. Financial benchmarks were another part of the roadmap, alongside payments and paid model access.
The proposal links funding with the cost of using an AI model: the route would put money with an agent, and a model call would incur a separate charge. Sharma's presentation cited x402 and Solana payment channels for machine payments, and Pay.sh for paid API access.
The plan sits alongside separate plans by Circle and Proof for a USD Coin (USDC) checkout demonstration for AI agents with buyer approval. EconoTimes previously reported that the proposed demonstration would record a buyer's signed approval before an agent pays.



