What will move agentic commerce from possibility to everyday reality?

Three things, and none of them is a better model. First, scoped authority the agent cannot exceed: a spending cap, approved merchants and a time window, set by the consumer and enforced at the network and issuer rather than by the agent's good behavior. Second, a receipt for the decision, not just the payment: what the agent was asked to do, what it considered, and why it chose what it chose. Third, the same protections consumers have today when something goes wrong. If disputing an agent's purchase is harder than disputing a human one, adoption stalls. Comfort will build outward from low-stakes, repeat purchases. Nobody starts by letting an agent buy a car.

The agent has to become genuinely useful. Consumers will delegate purchasing when an agent can consistently find what they actually need and match their preferences, constraints and context better than they could themselves — while saving them time. Controls such as spending limits, approvals and transparency will matter, particularly at the beginning. But we should not overestimate their role in adoption. The strongest incentive will simply be a better outcome with less effort. If the agent repeatedly gets the purchase right, trust and autonomy will follow.

Trust has to follow the customer’s intent through the entire purchase. The merchant needs to know the agent is authorized, the payment needs to stay within that authorization, and the product needs to match what the customer requested — what we call the "shopper's complete intent." If I ask for a replacement water filter, the right brand at the right price still fails if it doesn’t fit my refrigerator, my price range and all my other specifications, together that forms "shopper intent." Verifiable permissions plus intent and accurate product data have to work together, with a clear record of what happened and a way to resolve mistakes.

Reliability brings trust, which is the most important ingredient for agentic commerce. Until agents can connect consumer intent to accurate merchant data and a secure transaction, they will remain assistants rather than true commerce agents.

One of the biggest gaps is between how much an agent knows about the shopper and how little the merchant’s catalog tells the agent about the product — this adds friction to the AI shopper and product matching process. In one of my recent tests, an agent found the dog dental chews I wanted but selected the wrong size. Those details become consequential when the shopper increasingly delegates the purchase. The merchant’s product data needs to make sizes, formulations, compatibility and intended uses explicit, and shopper agents need to check those details against the request. Reliable autonomy depends on getting that match right consistently across millions of products.

Payment standards for agentic payments that everyone can adopt right now. It also means a severe course correction is going to be required for the absence of a standard today, where agentic payments are already happening. No single player should be able to dictate the future of agentic payments. And that standard that makes the capability available to everyone, large and small. At the same time, a player that is doing the wrong thing cannot be rewarded for it, allowed to capitalize on it, and build a position on top of it.

2026. The year agents went to work at scale and the personal agent arrived for everyone, not just developers. Workflows started running for days without a human in the loop and paying for services along the way. Nobody will point to a launch. They will point to the first boring overnight wins: work done, charges cleared, nothing to review. Once a provider had to be payable by an agent to be used at all, agents stopped being chat and became operators that could pay.

The tipping point will be when agentic commerce moves from isolated, closed experiences to an interoperable ecosystem. That means merchants can make their storefronts available across AI assistants, messaging platforms, publishers and other channels without rebuilding their commerce stack for each one. 

It will be the first time a consumer disputed a purchase an agent made and the whole chain, agent, merchant, acquirer and issuer, could show within minutes exactly what the agent was authorized to do and what it actually did. Payments infrastructure earns trust in failure, not in demos. If you want a date, the candidates are already behind us: September 2025, when [the agentic protocols] ACP and AP2 shipped in the same month, and June 2026, when Agent Pay for Machines let agents pay other agents for services. That second one changed the mental model from "AI helps you shop" to "AI is an economic participant." But the moment that will look decisive in hindsight is the first messy incident the system handled well.

The everyday, trivial, mundane tasks that nobody enjoys doing. People are clearly not asking AI to solve Olympiad-level problems for them everyday, but they simply want to hand off the grunt work. That is where habit forms, and it is why the payments have to be right from the very first transaction.

The boring one: replenishment. An agent that keeps the household stocked within a cap you set, from merchants you have approved, and sends a one-line summary of what it did. It will feel like a subscription that can think. It notices you are away next week, that the price of something has doubled, that a substitute is fine this time and not the next. The tell that it has become habitual will be that people read the receipt more carefully than they ever watched the purchase.

Commerce as maintenance, not a shopping trip. People already run their working week inside Claude and ChatGPT, and the habit spills into life: renew, switch, cancel, top up. Tools, media, data subscriptions, kept in line by an agent under a mandate and a tokenised card. No big checkout moment. The first weekly habit is the one you stop noticing, and it starts at work, where the call the agent pays for earns its cost back the same day.