There's a lot happening. In the last six weeks alone: Zuckerberg announced "agentic shopping tools" on a January earnings call, Reddit's COO used the phrase "agentic commerce" on a February earnings call, Alibaba ran 120 million orders through a chat interface in six days, and Google took the stage at NRF with Walmart's CEO to launch a universal commerce protocol. Microsoft was there too. Visa and Mastercard are building agent payment rails. Stripe partnered with OpenAI. Amazon is running its own shopping agent while simultaneously blocking external agents from accessing its listings.

That's all in the first 53 days of 2026.

Before getting into what it means, it's worth actually looking at what's being built — because the announcements are real and the underlying moves are bigger than the headlines.

What's Actually Getting Built

Meta is the one that surprised me most. Nobody in the agentic commerce conversation is properly tracking them. They're framed as an AI ads story, or an AI glasses story. But Zuckerberg used the exact phrase "agentic shopping tools" on the Q4 earnings call in January — describing agents that search Meta's advertiser catalog based on user history, interests, and relationships. That's the world's richest behavioral graph running in reverse. For 20 years Meta matched ads to people. Now they're flipping it: matching people's needs to products.

Then Susan Li, the CFO, said the Business AIs already live inside WhatsApp — currently at over 1 million weekly conversations in Mexico and the Philippines — are going to move from answering product questions to "help people get things done right within WhatsApp." That's transaction language. WhatsApp paid messaging crossed a $2 billion annual run rate in Q4, and the roadmap is to expand it to full transactions. They also acquired Manus in December for $2-3 billion — an autonomous AI agent with $100M+ ARR that can execute multi-step tasks end-to-end. That fills the last gap in the stack: they have the social graph, the product catalog, the messaging surface, and now an agent that can actually complete the purchase.

3.3 billion daily active users. $115-135 billion in CapEx planned for 2026. They are not hedging.

Reddit is already testing an AI-powered shopping search tool with select users in the US. Search for a product recommendation and you get interactive carousels — prices, product images, direct retailer links — drawn from recommendations people have shared in actual Reddit posts and comments. The results connect to Reddit's advertising and shopping partners, creating a direct line from community discussion to purchase. Weekly search users grew from 60 million to 80 million over the past year, and CEO Steve Huffman called AI search "a significant business opportunity."

But the strategic argument is what makes Reddit interesting. COO Jennifer Wong said on the Q4 earnings call: "The execution of [the last click] could be commoditized, frankly, because that's where the agent will get the job done. But the human, the last point of decision-making of what to buy and allocating resources, is at the recommendation, and Reddit has the best recommendations for products and services." She's saying: we're not the checkout layer, we're the trust layer. LLMs already cite Reddit in over 40% of responses. Reddit Answers grew from 1 million to 15 million weekly users. If agents are pulling product recommendations from somewhere, Reddit is positioning to be that somewhere.

China is not experimenting. While Western platforms are running pilots and announcing protocols, Alibaba ran the whole stack — discovery, comparison, checkout, payment — inside the Qwen chat interface during Chinese New Year. 120 million orders in six days starting February 6. Daily active users surged 727% on day one. Alipay's AI Pay hit 120 million transactions in a single week and 100 million total users over the holiday. They launched the Agentic Commerce Trust Protocol in January — an open framework connecting AI apps to commerce platforms and payment rails through a single integration.

The structural reason China can do this is straightforward: Alibaba owns the AI layer (Qwen), the commerce platform (Taobao/Tmall), the payment rails (Alipay), and the consumer app. They didn't need four companies to agree on a protocol. They shipped it internally and opened it to partners. ByteDance is right behind them — Doubao 2.0 launched February 14, already integrated with Douyin e-commerce, 155 million weekly active users, explicitly positioned as a model "for the agent era."

Google, Microsoft, Visa, Mastercard, Stripe are all at the infrastructure layer — building the pipes that would let any agent transact anywhere. At least six competing protocols are running simultaneously right now. IAB Tech Lab has flagged fragmentation risk. That's a polite way of saying the Western ecosystem is still figuring out how to agree.

The Half That Nobody Wants to Talk About

Here's the thing. Every major platform can build a commerce surface. Just like every major retailer could set up a retail media network.

Remember that period a few years ago when every retailer suddenly launched a retail media network? Target had one, Kroger had one, CVS had one, dozens of others followed. The announcements were real. The ad inventory was real. But Simon-Kucher is now literally asking "Are agents coming to kill retail media networks?" — implying that a large chunk of those networks never got meaningful adoption in the first place. Most of them aren't seeing the results that justified the investment. The infrastructure existed. The demand didn't materialize at scale.

That's the dynamic repeating right now with agentic commerce surfaces. The surfaces are real. The announcements are real. But several things haven't been solved, and they're not small things.

Data quality is the first one. 82% of executives name data quality as the biggest barrier to GenAI working. Product catalogs were built for SEO and human browsing — structured for keyword ranking and visual display, not machine reasoning. An agent trying to determine whether a product is right for a specific use case is working off catalog data that was designed to show up in search results, not to answer questions. That's a fundamental mismatch. You can't fix it by adding a commerce surface on top.

Adoption is stubbornly low. Only 2% of ChatGPT queries today are shopping-related. One-third of consumers say they'd be comfortable completing an agent purchase — which sounds promising until you notice that two-thirds won't. 71% of merchants report that AI merchandising tools have had limited to no effect. Forrester's headline for their 2026 predictions used the word "regrets." That's not a doom forecast, but it's a signal that the market is running ahead of real user behavior.

Trust is a real problem, not a messaging problem. It's one thing to use an AI to research a purchase. It's another to hand a card number to an agent and tell it to buy something on your behalf. The authorization frameworks are being built — Visa's Trusted Agent Protocol, Mastercard Agent Pay, Stripe's Agent Commerce Protocol — but they're infrastructure. Consumer trust is a behavior, and behavior changes slowly.

Fulfillment is still physical. Every agentic commerce surface has the same limitation: someone still has to pick, pack, and ship the order. The brilliant thing about the Alibaba example is that they own logistics too. In the West, that's not true for most of the platforms building commerce surfaces. Reddit is not shipping boxes. Meta is not running warehouses. The agent can complete the transaction, but the brand still has to deliver the product. That's not an argument against agentic commerce, but it's a reminder that the surface is not the hard part.

The Retail Media Network Parallel

When I look at this landscape, I keep coming back to retail media. The parallel is almost exact.

2020-2022: Every retailer announces a retail media network. The pitch is compelling — first-party purchase data, closed-loop attribution, advertising that actually converts. The announcements are credible. The infrastructure gets built. Budgets start flowing.

2024-2026: The fragmentation problem becomes obvious. Advertisers are managing a dozen different retail media platforms, each with its own interface, measurement methodology, and minimum spend. The mid-tier networks are seeing tepid adoption. The inventory exists but the demand is scattered. Only the platforms with massive reach — Amazon, Walmart, maybe Instacart — are seeing real scale.

Now look at agentic commerce surfaces. 2025-2026: Every major platform announces a commerce surface or protocol. The pitch is compelling — agents that can discover and transact at the point of intent. The announcements are credible. The infrastructure is being built.

The question is the same as it was for retail media: which surfaces will have real adoption? Not which ones will get built. They'll all get built. But which ones will consumers actually use? Which catalogs will be accurate enough for agent reasoning? Which payment rails will be trusted enough for autonomous transactions? Which fulfillment operations can actually back up the commerce claims?

Amazon is the only player that has a real answer to all of those questions simultaneously — which is probably why they're both running their own Buy for Me agent and blocking competitors like Perplexity Comet from accessing their listings. They know what the moat is.

What Actually Matters

The thing that happened in China matters more than any Western announcement right now. Not because 120 million orders in six days is a number we can transplant to the U.S. market — we can't, the structural conditions are completely different. But because it proves the model works when someone owns the full stack. Closed-loop systems ship faster than open-protocol ecosystems. That's not new information, but it's useful to see it demonstrated at transaction scale.

The Meta story matters because the distribution is real and the strategy is coherent. 3.3 billion daily active users is not a rounding error. If WhatsApp Business AI moves from answering product questions to completing transactions, that's a global commerce shift that will happen in markets most U.S. brands aren't paying attention to — Latin America, Southeast Asia, India — before it reaches domestic feeds.

Reddit's thesis is interesting and probably right. The recommendation layer will matter more than the execution layer. If agents are autonomous at checkout, the competitive advantage shifts upstream to whoever the agent trusts for product guidance. Reddit's bet is that they're already embedded in that position.

The protocol fragmentation is a problem that will work itself out, but slowly. Six competing agent payment protocols running simultaneously means the market hasn't picked its rails yet. Until it does, brands building for agentic commerce need to hedge across standards rather than bet on one.

And the data quality problem — the one that 82% of executives are pointing to — is the unglamorous work that determines whether any of this actually converts. A beautiful commerce surface sitting on top of incomplete, SEO-optimized, human-browsing-optimized catalog data is not going to perform. The brands that fix their data infrastructure now will have a real advantage when the surface wars settle.

Everyone can announce a commerce surface. The question is what's behind it.

The Invisibility Arc

In Instant Checkout, I write about what I call the Invisibility Principle — the pattern where truly transformative technology follows a predictable arc: novelty, utility, integration, invisibility. Electricity did it. GPS did it. The internet did it. Commerce will too.

What we're watching right now — Meta, Reddit, Alibaba, Google, Microsoft, all racing to make their surface shoppable through agents — is the messy middle of that arc. We're somewhere between utility and integration. The technology works. The announcements are real. But the seams are showing everywhere. Six competing protocols. Catalog data built for humans, not machines. Trust frameworks that exist on paper but not in consumer behavior. Fulfillment that still requires someone to put a product in a box.

The surfaces will keep multiplying. That's the easy part. The hard part — the part that determines who actually wins — is the invisible infrastructure underneath: clean data, earned trust, reliable fulfillment. The platforms that solve those problems won't need to announce anything. Their commerce will just work. And eventually, like electricity, nobody will think about it at all.

The ones that skip that work and just slap a buy button on an AI chat window? They'll be the retail media networks nobody uses.