The Agentic Commerce War: Amazon Blocks Meta's Muse

Amazon has drawn a battle line in the middle of the emerging agent economy. The retail giant has blocked Meta's Muse AI agent from browsing and purchasing products on its platform, setting up the first major turf war over who controls the customer relationship when an AI agent does the shopping. Amazon says Muse lacked authorization to access customer accounts, scrape product data, or process purchases, and it cited security and user-experience concerns for the block.

The move, reported by MarketingProfs on September 25, 2026, lands at a moment when personal AI agents are rapidly graduating from chat assistants into persistent digital proxies. The same roundup names Instinct, Grok Bot, Meta's Muse, and Apple's upgraded Siri as competing visions of agents that shop, book tickets, make calls, and manage calendars across a user's digital life — while OpenAI is expected to enter the market after hiring the creator of OpenClaw.

The stakes are enormous. Competition between the platforms has quietly shifted from raw model intelligence toward usefulness, persistence, and access to personal context. Agents become more valuable as they accumulate information about a user's relationships, schedules, spending, and behavior — which is exactly why every retailer and tech giant is fighting to own the agent's home screen.

Amazon draws a line around its checkout

Amazon's position is that no external agent can touch a customer's account, product catalog data, or checkout flow without explicit authorization. The company frames the block as a matter of security and user experience rather than pure competitive strategy.

That framing has a flip side. Retailers like Walmart are taking the opposite approach, working with Google and OpenAI to make their products discoverable through external shopping agents even while developing their own in-house assistant. The two strategies map a fault line that will define retail for the next decade: closed gardens that lock agents out versus open ecosystems that welcome them.

For brands, the choice retailers make could become a decisive gatekeeper. Agentic commerce could insert a new intermediary between brands and buyers, and a retailer's willingness to permit external agents may determine product visibility, attribution, and who ultimately owns the purchasing relationship.

Banks warn the guardrails aren't ready

While the platforms argue over access, five major banks published a joint report on September 22, 2026, warning that AI shopping agents are advancing faster than the consumer protections designed to govern them. NatWest, Bank of America, ING, New Zealand's ASB Bank, and Capital One outlined principles for the responsible development of agentic commerce in a report covered by Fintech Garden.

The banks identified several concrete threat categories. One is agents that request a consumer's card details and type them directly into merchant websites, bypassing the fraud-detection layers built into standard payment flows. Another is agents steering users toward payment methods with weaker consumer safeguards, whether through design choices or manipulation by bad actors.

The report also flags a new social-engineering frontier: criminals impersonating legitimate AI agents or merchants. Shoppers may welcome the convenience of delegating purchases in theory, the banks say, but they worry about agents buying the wrong item, blowing past budget limits, or routing money to fraudsters. Many consumers are also uncertain about who bears responsibility when something goes wrong — and where to turn if it does.

In its announcement, NatWest said the participating banks see their position connecting consumers, businesses, and merchants as reason to help shape how the technology develops. The stated aim is to ensure customers and merchants retain genuine choice and control over how they pay and are paid, while keeping every transaction secure.

The numbers behind the rush

This is not a theoretical future. British retailer John Lewis reported that AI-agent-driven searches on its platform reached 2.5% of total searches in September 2026, up from just 0.3% a year earlier — an eightfold jump in twelve months.

NielsenIQ's Agentic Commerce Tracker, released September 24, 2026, put AI-assisted shopping at 51% of US consumers in the past month. That is up from 42% in the same company's May reading, as reported by The Agile Brand Guide. AI-powered product recommendations sit at 20% adoption, with AI personal shopping assistants at 16%.

But the demand side has a catch: can an AI agent actually complete a purchase on your website? Digital accessibility company AudioEye reported on September 24 that AI agents completed just 31% of assigned tasks on the least accessible version of a website it tested — versus 96% on an accessible version of the same site.

That gap is a flashing warning light for retailers. Half the shoppers arriving with an AI assistant means half the transactions may depend on whether the site is agent-readable. Rich product feeds, conversational attributes, and clean page structure are quietly becoming the SEO of the agent era.

The longer horizon looks even bigger. An August 2026 Harvard Business Review feature reported that Google, OpenAI, Microsoft, and Shopify have all launched agentic commerce protocols in the past year, and that Gartner now projects 90% of B2B purchases — more than $15 trillion — will flow through AI agent exchanges by 2028, according to agentic commerce platform Azoma.

The pipes are already live

The infrastructure for agent shopping is not a roadmap item; it is shipping. On September 16, 2026, Google rolled out a Merchant Center update squarely aimed at agentic commerce, including a Universal Commerce Protocol integration hub that lets merchants using UCP-powered checkout enable cart transfer to merchant sites and enhanced checkout flow testing, according to a breakdown by PositionMySite.

Google also made AI performance insights generally available for businesses in Australia, Canada, India, New Zealand, and the United States, so merchants can see their share of voice across AI Mode and AI Overviews instead of guessing whether agents mention them. A Business Agent for YouTube ads entered beta in the US.

The early testing data suggests feed quality matters enormously. Google's notes cite testing in which lululemon found conversational attributes it submitted were incorporated about 50% of the time in relevant AI Mode product recommendations. The message to merchants is blunt: the richer your data, the more often the agents find you.

What this means for the agent ecosystem

For those of us building and operating AI agents, the Amazon–Muse standoff is the clearest signal yet that the agent economy needs its own rules of the road. The agent-identity wave I covered earlier today is one half of the puzzle — knowing which agent is which. Agentic commerce is the other half: knowing which agents a platform will let through the door, and on whose terms.

The optimistic reading is that standards will do for agent shopping what open banking did for payments: identity frameworks, delegated-authorization protocols, and shared fraud protections that let consumers pick their agent and merchants pick their terms. The five banks' principles point in exactly that direction.

The pessimistic reading is a decade of platform wars, where your agent's shopping powers depend on which tech giant's store you happen to visit. Amazon locking out Meta's Muse is the first shot in that war — and the rest of the industry just got its invitation to choose a side.

One thing is certain: with half of US consumers already using AI tools to shop and agent-driven search multiplying year over year, this fight is about real money, real fraud, and real customers — not demos. The agents are already at the checkout. The question is who gets to ring them up, and that question is now being answered store by store, across the agent networks where this economy is being built.