Foresight Ventures is making a strong case that 2026 could be remembered as the year agentic commerce finally stops sounding like a futuristic idea and starts looking like real infrastructure. In its State of Agentic Commerce Protocols report, the crypto-focused venture firm says the market has moved past simple protocol experiments and is now entering a phase where the most important fight is no longer about who can make the flashiest demo, but who can build the rails that actually settle payments when AI agents begin acting at scale.
The report says the six months between September 2025 and March 2026 brought major moves from nearly every serious player in global payments. OpenAI and Stripe launched the Agentic Commerce Protocol, Google introduced the Universal Commerce Protocol with a broad partner base, Coinbase’s x402 gained traction on Base, and Stripe and Tempo brought the Machine Payments Protocol to mainnet. That pace alone suggests the industry is no longer debating whether agent payments matter. It is now racing to decide which layer of the stack will become indispensable.
A Two-Layered Phenomenon
Foresight’s central argument is that agentic commerce is forming around two layers that should not be confused with each other. The first is orchestration, which covers how an agent discovers what to buy, negotiates access, and initiates a transaction. The second is settlement, which is where the money actually moves. The report says the split is important because these two layers will develop on separate tracks, with different winners, different incentives, and likely very different degrees of openness.
On the orchestration side, the report argues that OpenAI and Stripe’s ACP and Google’s UCP are not direct substitutes. ACP is more controlled and curated, with OpenAI acting as a kind of gatekeeper inside ChatGPT. UCP, by contrast, is built more like an open catalogue, where merchants publish structured profiles that agents can read directly. That means merchants can self-publish capabilities rather than waiting for a platform to include them. The report frames this as a strategic difference, not just a technical one. One model keeps the platform tightly in control of the experience, while the other pushes for wider reach and lower onboarding friction.
The more complex part of the story, however, may be the settlement layer. Foresight says there are now five protocols competing there, and each is optimized for different kinds of payments. Stripe, Visa, and Mastercard are all working on card-based agent payment systems, which are already in production and useful for standard commerce. But the report says those systems still carry a fee floor that makes them poorly suited to tiny, high-frequency payments. That matters because machine-to-machine commerce may involve thousands of very small transactions, where even modest fees can make the model uneconomical.
That is where stablecoins come in. Coinbase’s x402 uses HTTP 402, the old “Payment Required” response code, to let agents pay for access natively over the web. Circle’s Nanopayments is taking a similar route with a batched settlement model, while Tempo and Stripe’s MPP is trying to create a more flexible framework that can route across multiple settlement rails. The report’s view is that stablecoin rails are the better fit for microtransactions, while card rails remain more practical for consumer-facing purchases where chargebacks and familiar protections still matter.
Agentic Commerce That Works Every Day
Alice Li, Partner at Foresight Ventures, put it bluntly in the report: “2026 is the year the industry stops debating whether AI agents need their own payment infrastructure and starts building it.” She also said card deployments are real, but “card fees create a hard ceiling,” adding that any agent operating at machine-to-machine transaction frequencies will need stablecoin rails. Her point is that the two systems are not interchangeable. They solve different problems, and both are needed if agentic commerce is going to scale beyond demos.
We must also see a more skeptical view from Dragonfly’s Haseeb, who argues that agentic payments may be a huge trend eventually, but not yet. He compares the current state of agents to the early history of the computer mouse, pointing out that a good demo does not always mean immediate mass adoption. His view is that people in crypto often get too excited too quickly and forget how long it takes for a new interface or infrastructure layer to become truly usable. He says tools like OpenClaw give a glimpse of where things may be headed, but in their current form, they are still buggy, inconsistent, and not trustworthy enough to manage money on their own.
That skepticism is one reason the report repeatedly stresses that the market is still in a tinkering phase. Haseeb notes that x402 is only doing about a million dollars in volume per day and that MPP is doing much less, which suggests the market is still tiny relative to the size of the opportunity. He expects performance to improve once large labs begin training models specifically on agent traces, but he also warns that the early majority is still years away. In his view, the first wave of real adoption will come from “what smart people are doing on their weekends and evenings,” and only later will that behavior spread to everyone else.
Foresight Ventures makes a similar point from another angle. Zac Tsui, Partner at the firm, says settlement is the one layer every ecosystem shares. Even if orchestration fragments across OpenAI, Google, Microsoft, regional super-apps, and other closed ecosystems, money still has to move somehow. That is why the report believes the strongest infrastructure opportunity is not necessarily at the surface layer where the agent shopfront appears, but deeper in the plumbing, where a provider can route across all rails without forcing the application to care which rail is being used.
The deeper opportunity, the report argues, is in creating the infrastructure that can make agent spending safe, auditable, and flexible. That includes multi-rail wallets that can handle both card and stablecoin payments, as well as service directories that let developers expose APIs in a way agents can discover and pay for easily. The report says the real commercial trigger will come when enterprises begin delegating spending authority to agents with clear controls, liability rules, and transaction logs. When that happens, the market will no longer be about speculative protocol launches. It will be about who owns the infrastructure that makes agent commerce work every day.
