Artificial intelligence. For decades, AI lived mostly in the public consciousness as science fiction fodder, the cliché macguffin. From HAL 9000 refusing to open the pod bay doors, Skynet sending robot assassins after teenagers or malevolent robots set on using humanity as batteries, and beyond, to the majority of people, AI was just a story. Then, slowly, it started sneaking into everyday life. Search engines got smarter. Phones began recognising voices and faces. Recommendation algorithms learned what we wanted to watch before we did. And now, with tools like LLMs, image generators, and autonomous agents, the idea of machines that can write, plan, code, create and make decisions no longer feels futuristic. It feels like the present.
Another important milestone in the evolution of AI is the ability for an agent to hold wallets, make payments, buy access to digital resources and distribute rewards without someone manually approving every step. That shift is what makes agentic payments so important. On Base, this is already moving from theory into live infrastructure, giving builders a practical way to let agents pay for data, API calls and user rewards in seconds.
What is an agentic payment?
An agentic payment is a transaction an AI agent makes autonomously, without a human clicking a button at the moment of purchase. The agent holds a wallet and keys and can makes decisions on what to buy, pays, and gets the resource back in seconds. Right now agents are in action at scale: they pay for data, API calls, make transactions to other agents and, in the case of Sliyd, hands out rewards to users based on their actions.
Agents break the established pattern. For most of software’s history, paying for something involved a person with a card at a checkout page, and perhaps most crucially the time to sit and make a payment.
What an AI agent needs to move value
Three things, and a card on file is not one of them.
First, a wallet the agent controls: an account it can sign from inside its own loop. A human’s wallet borrowed for a single transaction does not count. The agent needs to act autonomously.
Second, a payment standard that fits how agents work. Agents call APIs and tools thousands of times. Each call might be worth a fraction of a cent. Card processing fees of roughly 2.9% plus a fixed charge per transaction make anything under about a dollar uneconomical, so per-call billing on cards never worked. Stablecoin payments settled on-chain remove the fixed-fee floor, which is why micropayments are viable for the first time.
Third, a way to prove the agent is allowed to spend. A user grants an agent a budget and a scope. The agent operates inside it. The payment rail has to carry that permission with the transaction.
Three protocols now cover these needs, and they are converging rather than competing. Coinbase’s x402 uses the HTTP 402 “Payment Required” status code: an agent requests a paid resource, the server replies 402 with payment instructions, the agent signs a stablecoin transaction, attaches the proof, and retries. The whole cycle takes seconds and needs no login. Google’s Agent Payments Protocol (AP2), backed by 60-plus organisations including Mastercard, PayPal, and Adyen, handles the authorisation layer through signed mandates that record what a user asked for, what the agent assembled, and what gets charged. Stripe’s Machine Payments Protocol launched in March 2026 with session-based streaming payments and Stripe’s compliance stack built in.
These cover different parts of the same problem and stack together: AP2 records consent and x402 moves the money.
Why rewards are the first agentic use case
Most agentic-commerce coverage points at shopping: an agent buying flights, groceries, or compute. Rewards distribution is the quieter, faster-moving case, and it fits agents better than retail does.
A reward is small, frequent, and rule-driven. A brand running a campaign wants to send a digital reward every time a user completes an action, hundreds or thousands of times a day, each one worth very little. That is the exact shape of transaction that broke on card rails and works on agentic ones. The decision logic (“did this user qualify?”) is something an agent already does well, and the payout is small enough that on-chain settlement is the only economic option.
The market data backs the direction. By June 2026, Chainalysis reported that x402 had crossed 100 million agentic transactions on Base, with Base’s own figures citing roughly 165 million settled transactions across more than 480,000 agents. The composition shifted too: transactions of a dollar or more grew to 95% of volume, up from 49% in early 2025, as the activity moved from speculative micro-spends toward real payments. That shift, from test traffic to dollar-scale settlement, shows agents moving value that matters on Base.
Rewards sit at the front of this curve because they need what agentic payments provide: programmatic decisions, tiny amounts, instant settlement, no checkout.
What’s live on Base today
Builders can wire agentic rewards into a product now. The pattern looks like this: an agent watches for a trigger (a completed action, a purchase, a streak), checks the campaign rules, and distributes a digital reward to the user’s wallet, settling on Base.
Sliyd runs a live MCP-compatible payment endpoint on Base for this. An AI agent connecting over the Model Context Protocol can call the endpoint as a tool, pay per call where required, and distribute branded rewards in code. Because it speaks MCP, any agent framework that already uses MCP tools can add reward distribution without bespoke integration work. The reward lands in the user’s wallet with no app download and no manual claim step.
This is the piece the broader market is missing. x402, AP2, and MPP define how agents pay. Far fewer projects define what agents pay for in a way a brand can use today. A live endpoint that turns “an agent decided a user earned something” into “the user has it” is the practical layer on top of the payment standards.
Where agentic payments go next
Agentic payments stopped being a theory in 2025 and became infrastructure in 2026. The standards are settling, the volume is real and growing, and the protocols are consolidating under neutral homes (Google is donating AP2 to the FIDO Alliance, and x402 now sits under the Linux Foundation with backing from Google, AWS, and others). For builders, the open question is no longer whether agents can pay. It is what they should pay for first. Rewards are a strong answer, and the rails to deliver them are live on Base.
FAQ
A payment an AI agent initiates and settles on its own, inside a task, without a human approving each transaction at the moment it happens. The user sets the budget and rules in advance; the agent operates inside them.
It’s not technically a “can’t” it’s more of a “don’t”. Card processing carries a fixed fee per transaction (around 2.9% plus a flat charge), which makes payments under a dollar uneconomical. Agents often transact in tiny micropayments at high frequency, so they need a payment option without a fixed-fee floor.
x402 is an open payment standard from Coinbase that uses the HTTP 402 “Payment Required” status code. x402 enables a server to ask for payment in response to a request; the agent can fulfill this payment in stablecoins. There have been more than 100 million x402 transactions on Base.
Yes, Sliyd can process x402 payments through its MCP support.
They cover different layers. AP2 records authorisation and consent through signed mandates. x402 moves the payment over HTTP. Stripe’s Machine Payments Protocol offers session-based streaming payments with built-in compliance. They stack together rather than compete.
An agent checks a campaign’s rules, decides whether a user qualified, and sends a digital reward to the user’s wallet, settling on Base. Sliyd exploits this through a live MCP-compatible endpoint, so any MCP-enabled agent can distribute branded rewards without a custom integration.
The Model Context Protocol is a standard way for AI agents to discover and call external tools. An endpoint that supports MCP can be used by any agent framework that already supports MCP, removing the integration work required to add a new capability such as reward distribution.
