Open Agentic Commerce: How AI Agents Will Buy in 2026

In September 2025, OpenAI and Google shipped competing "open" payment standards in the same two weeks. By the 2025 holidays, AI and agents influenced roughly 20% of US retail sales, about $262 billion, according to Salesforce. The agent that picks your flight or refills your protein powder isn't a someday story. It's already spending money, and the rails it runs on are being poured right now.

Most explainers on this topic are written for merchants worried about getting their store "agent-ready." This one takes the other seat. Agentic commerce is really a fight over who owns the checkout button, and "open" is the word deciding whether your agent can buy anywhere or only where someone paid for the privilege. That distinction matters more than any single product launch.

This post breaks down what open agentic commerce actually is, how the new protocols move money, who is building them, and what it means for you whether you're shopping, selling, or building agents. You'll leave knowing the difference between a real open standard and a walled garden wearing an open badge.

TL;DR

• Agentic commerce is online buying where an AI agent discovers, compares, and completes a purchase on your behalf, not just suggests options you click yourself.
• "Open" means the standards are public and payment-agnostic, so any agent can transact with any merchant through any payment provider, instead of being locked into one company's checkout.
• Two open protocols launched within weeks in September 2025: OpenAI and Stripe's Agentic Commerce Protocol (ACP) and Google's Agent Payments Protocol (AP2).
• The money forecasts are large: McKinsey projects $3 to 5 trillion globally by 2030, and Gartner expects AI agents to intermediate over $15 trillion in B2B spending by 2028.
• The contrarian catch: the protocols are open, but the front doors aren't. Real openness lives in the agent you bring, not the checkout you rent.

What Is Open Agentic Commerce?

Open agentic commerce is online buying where an AI agent acts on your behalf to find, compare, and pay for things, using open protocols that aren't owned by a single platform. The "agentic" part means the agent executes the purchase. The "open" part means it can do that across the whole internet, not just inside one company's app.

Strip out the buzzwords and it's a delegation shift. You stop being the person who clicks "buy." You become the person who tells an agent what you want and sets the limits. IBM frames agentic commerce as agents that handle discovery, execution, and post-purchase support within guardrails you define. The agent proposes. You set the rules. The transaction happens without you babysitting every step.

Conversational vs autonomous commerce

These two get blurred constantly, and the difference is the whole game. Conversational commerce is chat as an interface: an assistant shows you three blenders, and you still tap the checkout button yourself. It improves your decision. Autonomous commerce is the agent actually completing the purchase inside the permissions you granted. It improves the outcome, because the work gets done, not just teed up.

Picture a recurring order. With conversational commerce, the bot reminds you that you're low on coffee and links the product page. With autonomous agentic commerce, the agent reorders your usual bag, confirms it fits your budget rule, and you find out when it ships. One is a smarter search box. The other is a delegate.

Open standards vs walled gardens

Here's where "open" earns its keep. An open standard is publicly specified and payment-agnostic, so any compliant agent can buy from any compliant merchant through any supported payment rail. A walled garden is the opposite: one platform owns discovery, checkout, and payment, and your agent only works where that platform already has a deal. Open is a public road. A walled garden is a private mall with one entrance.

This is the real fork in agentic commerce. If the standards stay open, your agent shops the open web. If they don't, you're back to ten apps and ten logins, except now the apps are agents.

How Does Agentic Commerce Actually Work?

An agentic purchase moves through four steps: intent, discovery, authorization, and payment. You state a goal. The agent finds candidates. It gets your signed permission to buy a specific thing at a specific price. Then it triggers payment through a rail that knows an agent is involved. Each step now has an emerging open standard behind it.

The hard problem was never search. It was trust. How does a merchant know an agent is authorized to spend your money, and how do you make sure it can't go off-script? Stripe's answer is a Shared Payment Token, a credential scoped to one merchant and one cart total, so the agent can pay without ever touching your raw card number. The card data stays out of the model's hands.

Mandates and tokens

Google's protocol leans on cryptographically signed "mandates." An Intent Mandate captures what you asked for. A Cart Mandate is a tamper-proof record of the exact items and price you approved. That signed chain is how AP2 separates authorization from the payment rail, so any network can verify the agent had permission. The agent can't quietly swap a $40 cart for a $400 one, because the signature wouldn't match.

The card networks are wiring up too. Mastercard launched Agent Pay with agent-aware tokens and completed its first live agentic transaction in late September 2025. Visa shipped its own intelligent commerce program in 2025. The plumbing is real, and it's shipping.

The Protocol War: ACP vs AP2

Two open protocols now dominate the conversation, and they landed almost simultaneously. OpenAI and Stripe released the Agentic Commerce Protocol. Google released the Agent Payments Protocol. Both are open source. Both want to be the default. Understanding the split tells you where agentic commerce is heading.

ACP: OpenAI and Stripe

ACP is the standard behind Instant Checkout in ChatGPT. Released in late September 2025 under an Apache 2.0 license, it connects buyers, their agents, and businesses to complete a purchase inside the chat. It launched with US Etsy sellers and a pipeline of over a million Shopify merchants. If you've bought something without leaving ChatGPT, you've used it.

The pitch is reach. ChatGPT has hundreds of millions of users, and ACP turns that audience into a storefront. Stripe handles the money through its Shared Payment Token, and the buyer never leaves the conversation.

AP2: Google and 60-plus partners

AP2 took the coalition route. Google announced it in mid-September 2025 with more than 60 partners, including Mastercard, American Express, PayPal, and Coinbase. It's payment-agnostic by design and built as an extension of Google's Agent2Agent protocol, with an add-on for stablecoin payments. In 2026 it was donated to the FIDO Alliance for neutral governance.

The bet here is interoperability. Rather than own the checkout, AP2 wants to be the signing-and-verification layer everyone else plugs into. It's the difference between owning the mall and paving the road every mall connects to.

Where MCP fits

Underneath both sits the Model Context Protocol, the open standard for how agents reach external tools and data. Think of the layers like this: MCP is how an agent gets the tools, and ACP or AP2 is how it pays once it has them. An agentic commerce protocol handles the transaction, while MCP handles the connection. A capable agent platform needs both, which is why the tool layer matters as much as the payment layer.

Why "Open" Is the Word That Matters

Open is the only thing standing between you and a re-run of the app-store era. If the standards powering agentic commerce stay genuinely open, your agent transacts with any merchant through any payment provider. If they don't, you're renting a shopping bot from whoever owns the checkout, and your buying power gets routed through their funnel and their fees.

Here's the contrarian part. We don't actually have "open" agentic commerce yet. We have three competing "open standards" racing to become the default, and the real lock-in is sliding from the checkout page to the agent's tool layer. A buyer inside ChatGPT's Instant Checkout is still inside OpenAI's funnel. AP2 still rides Google's stack. The specs are open. The front doors are not.

The genuinely open position is the independent agent: one that speaks every protocol and isn't owned by a marketplace, a chat app, or a card network. That's the whole thesis behind a platform like Poncho, where one account reaches 3000-plus tools and the agent works across the open web instead of one company's storefront. Bring your own agent to the internet. Don't rent one tied to a single checkout button.

What Open Agentic Commerce Means for You

Your move depends on which seat you're in: buyer, builder, or seller. For buyers, the win is delegation. For builders, it's a new surface to build on. For sellers, it's a discovery channel that doesn't care about your homepage. All three are arriving faster than most roadmaps assume.

The momentum is hard to argue with. Morgan Stanley estimates agentic shoppers could drive $190 to $385 billion in US e-commerce by 2030, with roughly 23% of Americans already making a purchase using AI in a recent month. Adobe clocked traffic to US retail sites from generative-AI sources up 693% year over year in the 2025 holiday season. The behavior shift is already in the numbers.

If you're a buyer

Start delegating the boring purchases. Replenishment, price-tracking, and research-then-buy tasks are where agents earn trust fastest. Set hard limits up front, a budget cap and an approval threshold, and let the agent run inside them. The skill to build now is writing a clear instruction, not clicking faster. If you want a feel for it, our guide on getting started with Poncho walks through running your first agent task in one sentence.

If you're a seller

Get your product data machine-readable and your inventory honest. Agents pick from structured feeds, not pretty hero images, so the brand that wins is the one an agent can parse and trust. Discovery is moving from "rank on the homepage" to "be the answer the agent returns." That's a different optimization problem, and the early movers are already working it.

Bottom Line

Open agentic commerce is the shift from clicking "buy" to delegating the purchase, riding on open protocols that decide whether your agent shops everywhere or nowhere. The standards arrived in 2025, the money is following on a multi-trillion-dollar curve, and the only real question is whether "open" stays open or quietly hardens into a new set of walls. Bet on the agent you control, not the checkout you borrow. The cleanest way to feel where this is going is to hand a real task to an agent that already speaks across thousands of tools and watch it run. Try it on Poncho and see what delegating a purchase actually feels like.

Frequently Asked Questions

What is open agentic commerce in simple terms?

It's online shopping where an AI agent does the buying for you, using public standards that work across many merchants instead of one closed platform. You describe what you want and set limits. The agent finds it, confirms the price against your rules, and completes the purchase. "Open" is the part that lets it shop the whole web rather than a single company's storefront.

How does an agentic purchase stay secure?

The new protocols keep your raw payment details away from the agent and require signed permission for every buy. Stripe's approach issues a token scoped to one merchant and one cart total, so the agent can pay without seeing your card number. Google's uses cryptographically signed mandates that lock in the exact items and price you approved. If the agent tried to change the order, the signature would break and the payment would fail.

What is the agentic commerce protocol?

It's an open standard for how a buyer, their AI agent, and a business complete a purchase together. The best-known one, ACP, was released by OpenAI and Stripe in September 2025 and powers Instant Checkout in ChatGPT. Google's AP2 is a parallel open standard focused on payment authorization across many providers. Both are open source, and both are competing to become the default rails.

Is agentic commerce the same as conversational commerce?

No, and the difference is who completes the sale. Conversational commerce uses chat to help you decide, but you still click buy yourself. Agentic commerce has the agent actually execute the purchase within the permissions you set. One is a smarter search and recommendation layer. The other is true delegation, where the task gets finished without you in the loop for every step.

How do brands get discovered by AI agents?

By making product information clean, structured, and machine-readable, because agents pick from data feeds rather than visual web pages. Accurate inventory, clear pricing, and structured product details matter more than homepage design when the shopper is software. As more buying runs through agents, discovery shifts from ranking on a page to being the answer an agent returns. Treat your product data as the new storefront.

Will AI agents really handle a meaningful share of buying?

The forecasts and early data both point that way. McKinsey projects agentic commerce could generate $3 to 5 trillion globally by 2030, and Morgan Stanley pegs US agentic shopping at up to $385 billion by 2030. Adobe already measured a 693% year-over-year jump in retail traffic from generative-AI sources over the 2025 holidays. The behavior is scaling, not hypothetical.

How do I start using agentic commerce today?

Pick a low-stakes, repetitive purchase and delegate it to an agent with clear limits. Replenishment orders, price-tracking, and research-then-buy tasks are the easiest places to build trust. Set a budget cap and an approval point, then expand once the agent proves reliable. A general AI-agent platform like Poncho lets you describe the outcome in plain English and have it pull from thousands of tools to get it done.