Google is testing a feature that lets you buy things from Flipkart without leaving Gemini. Google is also a company whose assistant products have a long record of being quietly retired, rebranded, or half-shipped. Both of those things are true at the same time, and that tension is the whole story here.
The facts are thin, and I want to be upfront about that. Google is running a limited test in India that allows purchases from Walmart-owned Flipkart through Gemini and AI Mode. It covers select products and select users. A broader rollout is planned for later in October. Flipkart, for its part, is pushing an AI-first commerce strategy, with a search bar that is becoming conversational and personalized, according to Chief Product and Technology Officer Balaji Thiagarajan. Google has previously described this kind of agent-to-retailer interaction as an open standard, meant to let AI agents work with retailers across the shopping journey, checkout included.
That’s it. That’s the whole verified pile. Everything else floating around this story is speculation, and I’d rather give you a useful read on what it means than pad it with invented numbers.
Why the checkout step is the part that matters
Most AI shopping features so far have been elaborate ways of not buying anything. You ask a chatbot for running shoes, it gives you a tidy list, and then it hands you off to a browser tab where you do the actual work. The assistant was a search box with better manners.
Completing a purchase inside the assistant is a different category of problem. It means handling payment, address selection, order confirmation, and — the part nobody demos — returns, cancellations, and the moment something goes wrong. A recommendation that misses is a mild annoyance. A purchase that misses is a support ticket, a refund, and a customer who now distrusts the whole interface.
So the interesting question isn’t whether Gemini can find you a phone case. It’s whether Google and Flipkart have built the unglamorous plumbing behind the confirmation screen. A limited test on select products and users suggests they know that too. Restricting the catalog is how you keep the edge cases manageable while you figure out what breaks.
India is a deliberate choice, not a coincidence
Running this in India first is a reasonable read on where the volume is. Flipkart is one of the largest e-commerce operations in the country, and the AI-first direction Thiagarajan describes lines up with Google needing a retail partner that actually wants an agent poking at its catalog. Plenty of retailers do not. An agent that answers product questions and completes checkout sits directly between a store and its customer relationship, which is not a position most merchants hand over cheerfully.
That’s why the open standard framing is the piece I’d watch most closely. If agent-driven checkout only works through bespoke deals between giant companies, it’s a partnership, not a standard. If smaller retailers can plug in without negotiating individually with Google, it becomes infrastructure. Those are very different outcomes, and the current test doesn’t tell us which one we’re getting.
What I’d want to know before calling this useful
From a toolkit reviewer’s seat, a feature is only worth recommending when you can answer basic operational questions. Here’s what’s still open:
- Who owns the order when it goes sideways — Google, Flipkart, or a pass-the-parcel between the two?
- How does price and stock accuracy hold up when the assistant is quoting from a cached view of a catalog?
- Can you actually see what you’re agreeing to before you confirm, or does the conversational layer hide the details that matter?
- Does the personalization make recommendations better, or just better at steering you toward higher-margin inventory?
- What happens to comparison shopping when one retailer sits inside the assistant and others don’t?
None of that is answerable from a limited test announcement. It’s answerable in about six weeks of real users placing real orders, which is presumably what the October expansion is for.
My honest take
I’m cautiously interested rather than impressed. The direction is right — assistants that can finish a task beat assistants that describe one — and picking a single large partner in a single large market is the sensible way to start. But conversational commerce has a long history of demoing beautifully and holding up poorly, and buying things is exactly where small failures get expensive.
If the October rollout arrives with clear order handling, honest pricing, and a path for retailers beyond one Walmart subsidiary, this turns into something worth building habits around. If it ships as a polished shortcut that quietly routes you to one store, it’s a distribution deal wearing an assistant costume. Worth testing when it lands. Not worth changing how you shop yet.
🕒 Published: