Think about the last time a friend recommended a restaurant and then, without asking, walked you to the door, pulled out your wallet, and handed the cash to the host. Helpful? Technically. Comfortable? Depends entirely on how much you trust that friend.
That is roughly what Google is testing in India right now. A limited test lets people buy from Walmart-owned Flipkart directly through Gemini and AI Mode in Search, with Flipkart-branded checkout showing up inside the flow. It covers select products and select users, and a broader rollout is planned for later in October.
That is the whole verified story. Short. But the shape of it matters more than the size.
What is actually new here
Shopping links in chatbots are not new. Product carousels in Search are not new. What changes with this test is the last step. Instead of an AI assistant handing you off to a website and wishing you luck, the checkout surfaces inside the assistant. The Flipkart branding staying visible in that flow is the detail I keep circling back to, because it tells you something about the negotiation behind it.
Retailers have spent a decade fighting to own the checkout moment. It is where the payment data lives, where the upsell happens, where the customer relationship gets cemented. Handing that step to a Google surface is a real concession. Keeping your logo on it is the compromise that makes the concession survivable.
Google has also described its agentic shopping work as an open standard meant to let AI agents interact with retailers across the shopping journey, checkout included. Read that as a land grab dressed in the language of interoperability. Open standards are wonderful. They are also a very effective way to make sure everyone builds toward your surface.
Why India, and why now
Testing this in India first is not a random choice. It is one of the largest mobile-first commerce markets on the planet, and Flipkart is one of its anchors. If agent-driven buying works anywhere, it works where people already do most of their shopping on a phone with a single dominant search app installed by default.
There is also less legacy friction. In markets where desktop shopping habits calcified over twenty years, asking someone to complete a purchase inside a chat interface feels alien. In a market where the phone is the store, it feels like one less tap.
My honest read as a reviewer
I test AI agents for a living, and the pattern is depressingly consistent. Agents are good at narrowing options. They are mediocre at the specific, boring details that determine whether a purchase was correct. Wrong variant. Wrong size. Wrong seller among five identical listings at different prices. Stale stock counts. A return policy that only exists on the fine-print page the agent skipped.
Those failures are annoying when an agent is drafting an email. They cost money when the agent has your payment method.
So the questions I want answered before the October rollout are not about the demo. They are about the plumbing:
- Who owns the refund when the agent picks the wrong item? Google, Flipkart, or the shopper who typed a vague sentence?
- Is there a confirmation screen showing the exact SKU, seller, and price, or does the assistant summarize and proceed?
- How does product ranking work? If a retailer pays for placement inside an AI answer, the user has no obvious way to tell.
- What happens to price comparison? Half the value of online shopping in India is cross-checking three apps. A single integrated checkout quietly removes that habit.
That last one is my real concern. Convenience and comparison are in direct tension. Every step you remove from a purchase also removes a moment where the buyer might reconsider. Retail has understood this since the invention of the one-click button. An AI assistant that researches, recommends, and charges you in one continuous motion is the most frictionless funnel ever built, and frictionless is not always the same as good for you.
What I would watch for in October
The test being limited to select products and users is the right call, and it is also a tell. You restrict scope when you are unsure how the edge cases behave. Watch which product categories expand first. If it stays in simple, low-variance goods, that tells you the agent still struggles with anything requiring judgment. If apparel and electronics show up early, either the systems got better or someone got brave.
I want this to work. Shopping across six tabs is genuinely tedious, and a competent assistant would be a real improvement. But competence here means being right about specifics, not fluent about generalities. Those are different skills, and the industry keeps shipping the second one while promising the first.
The checkout button is the honesty test. We will see how it scores.
🕒 Published: