Japan's AI Agents Could Skip Amazon — And Rewire Asian E-Commerce

The Disintermediation No One Was Predicting

Amazon built its empire on a simple transaction: merchants pay commissions and advertising fees to reach shoppers, and Amazon controls the discovery layer. That model is now colliding with something it never accounted for — AI agents that don’t need a marketplace to find products.

LINE Yahoo announced in late August that it will launch “Agent i,” a standalone AI shopping agent app, by October. The service lets users chat in Japanese, upload images of clothing or household items, and receive direct product recommendations pulled from Yahoo Shopping and — crucially — from rival and independent sites. It will also plan travel itineraries, aggregate calendar availability, and offer pet-health consultations.

This sounds like a feature upgrade. It isn’t. It’s a platform-level pivot that could hollow out Amazon’s third-party seller ecosystem across Asia.

Amazon Is Falling Behind on AI — and It Shows

According to the source reporting, Amazon has stumbled over AI implementation. Its strategy relied on external foundation models, and licensing costs came in higher than expected. The company reportedly struggles to integrate generative-AI capabilities into its existing retail stack. Meanwhile, competitors are shipping.

LINE Yahoo disclosed that AI-agent-driven transactions on Yahoo Shopping reached 20% of total volume by July — up rapidly from the February launch of its earlier agent features. Near 40% of users engage with recommendation functions like price-drop alerts and side-by-side comparisons. The company is now expanding agents to source across multiple retailers, not just its own storefronts.

Rakuten is moving in parallel. It plans a beta “AI Store Manager” inside its marketplace by year-end — an agent that learns each merchant’s personality and selling style, then handles customer interactions 24/7. That’s a natural fit for Rakuten’s tenant-model architecture, where hundreds of thousands of independent shops operate under one roof.

Amazon, by contrast, is still trying to figure out how to make its existing model AI-compatible without cannibalizing the fees that fund it.

The “Amazon Bypass” Is Already Happening

The Japanese press is calling it “Amazon toss” — shoppers and agents routing around Amazon entirely. The mechanism is straightforward.

An AI agent doesn’t inherit loyalty to any single platform. It inherits instructions. When a user says, “Find me a lightweight jacket under ¥5,000 for a trip to Southeast Asia,” the agent queries product databases, reviews, and pricing across multiple retailers. If a direct-from-manufacturer site or a smaller web store offers a better margin or a better product, the agent recommends it. The shopper buys outside Amazon. The commission evaporates.

This is especially dangerous for Amazon because it attacks two revenue streams at once:

• Seller fees. Third-party merchants pay referral commissions on every sale. Fewer sales through Amazon means less fee income.
• Advertising revenue. Amazon’s most profitable segment isn’t retail — it’s sponsored product listings. Merchants bid to appear at the top of search results. If AI agents distribute products across open links rather than Amazon’s paid slots, that bidding market shrinks.

The math favors bypass. An AI agent’s inference cost per query is measured in cents. A seller’s Amazon ad spend can run into thousands of dollars monthly. When agents lower customer-acquisition costs below marketplace fees, merchants have financial incentive to shift traffic elsewhere.

What This Means for Platform Capitalism in Asia

The Amazon model depends on scale. More sellers attract more buyers, which attracts more sellers. It’s a classic two-sided platform loop. AI agents break the loop by making discovery independent of platform size.

In Asia, where mobile-first commerce is already dominant and super-app ecosystems like LINE and Rakuten bundle messaging, shopping, finance, and media, the agent architecture fits naturally. LINE Yahoo isn’t building a separate store — it’s building an intelligent gateway that sits above the marketplaces and decides where traffic flows. That shifts power from the platform that hosts sellers to the platform that controls attention.

Rakuten’s AI Store Manager takes this further. Instead of centralizing discovery, it decentralizes the selling experience itself — giving each merchant an AI copy of their best salesperson. The marketplace becomes infrastructure, not destination.

Amazon’s response will determine whether it remains the default checkout page in Asia or becomes one option among many.

Who Wins, Who Loses

Merchants. Small and mid-size brands win if agents route them directly to consumers. They shed advertising overhead and retain customer relationships. Large sellers with massive Amazon ad budgets may resist, but the economics increasingly favor disintermediation.

LINE Yahoo and Rakuten. They become the gatekeepers of intent rather than the gatekeepers of inventory. Agents create stickier relationships — users return to the tool that learns their preferences, not the catalog that happens to have stock.

Amazon. It faces a dual threat: declining marketplace GMV and a collapsing advertising moat. Its scale advantage weakens when agents ignore platform boundaries. The company’s delayed AI rollout makes recovery harder.

Consumers. They get cheaper discovery and more choice. But they also hand personal preference data to whoever operates the agent — a tradeoff that will generate regulatory scrutiny in the coming years.

What Happens Next

Expect agent-driven commerce to accelerate through Q4 2026 as LINE Yahoo’s Agent i launches and Rakuten’s store managers enter beta. The first measurable signal will be Amazon Japan’s advertising-revenue growth rate. If it stalls or reverses while overall e-commerce expands, the bypass thesis is confirmed.

Regulators in Japan and South Korea may also take notice. AI agents that aggregate pricing and redirect traffic sit in a gray area between search and intermediary liability. How platforms handle consumer data — especially health queries from pet-care agents or travel preferences — will shape the legal landscape.

The deeper story isn’t about a new app. It’s about who controls the moment before the purchase. Amazon spent two decades owning that moment. AI agents are giving it away.

The Uncomfortable Truth for Western Readers

American coverage of Asian tech tends to focus on valuation stories or hardware launches. The real shift in Japan right now is structural and quiet — platforms rewriting the rules of e-commerce from the inside out. LINE Yahoo and Rakuten aren’t copying Amazon. They’re making Amazon irrelevant by changing the question from “where do I shop?” to “what should I buy?”

The agent answers the second question. The marketplace loses the first.