Remember when a search engine’s entire job was to point at things? You typed “running shoes,” you got a page of links, and the actual shopping happened somewhere else entirely. Google handed you the map. You did the driving, the comparing, the cart-filling, and the card-entering.
That arrangement is quietly being renegotiated. Google is testing a setup in India where you can buy products from Flipkart, the Walmart-owned retail giant, directly through Gemini and AI Mode. Right now it’s a limited test covering select products and select users, with a broader rollout planned for later in October.
Small news on the surface. Fairly big shift underneath.
What’s actually changing here
For anyone who doesn’t spend their days reading AI release notes, here’s the practical difference.
Today, if you ask an AI assistant about a product, it acts like a very well-read friend. It explains, compares, maybe suggests. Then it taps out and sends you to a website to finish the job.
What Google is testing moves the assistant from advisor to errand-runner. You ask, it finds, and the purchase can happen without you bouncing out to a separate storefront and repeating your whole thought process into a search bar.
That’s the line between a chatbot and an agent. A chatbot talks about the thing. An agent does the thing. Checkout is the moment where that distinction stops being philosophical and starts involving your money.
Why Flipkart is an interesting dance partner
Flipkart isn’t being dragged into this. The company has been moving toward an AI-first commerce strategy of its own, turning its search bar into something conversational and personalized rather than a keyword slot. Chief Product and Technology Officer Balaji Thiagarajan has spoken about that shift directly.
So you have two parties arriving at the same idea from opposite ends. Google wants its assistant to complete tasks instead of handing them off. Flipkart wants shopping to feel like a conversation instead of a filter menu. A test that connects Gemini to Flipkart’s catalog is where those two ambitions overlap.
It also matters that this is happening in India first. That’s one of the largest and most competitive online retail markets anywhere, and it’s mobile-heavy, price-sensitive, and full of shoppers who compare obsessively before committing. If agent-driven buying can earn trust there, it can probably earn trust most places.
The plumbing nobody sees
None of this works on vibes. For an assistant to buy something on your behalf, there has to be an agreed-upon way for AI agents to talk to retailers across the whole shopping journey, checkout included. Google has described an open standard built for exactly that, allowing shoppers to buy eligible items through an agent rather than a browser tab.
Think of it as a shared language between the assistant and the store. Without it, every retailer integration is a one-off custom job. With it, the assistant can theoretically work with many stores using the same set of moves.
That’s the part worth watching, because standards are how these things stop being demos and start being infrastructure.
What this means if you’re not a developer
A few honest expectations, based on how these rollouts usually go:
- Limited means limited. Select products, select users. If you’re in India and don’t see it, that’s normal.
- You’ll still be in the loop. Early agent commerce tends to keep confirmation steps in place. The assistant assembles the order; you approve it.
- Convenience has a tradeoff. When one assistant becomes the front door to shopping, the question of which products it surfaces, and why, gets a lot more consequential.
- Trust gets built on boring things. Correct sizes, correct addresses, clean refunds. That’s what decides whether people use this twice.
The part I find genuinely interesting
We’ve spent a couple of years watching AI assistants get better at explaining, summarizing, and drafting. Those are all low-stakes tasks. If a summary is mediocre, you shrug and rewrite it.
Buying is different. A wrong purchase costs real money and real time. So when a company starts wiring an assistant into checkout, it’s making a statement about how much it trusts its own system to act rather than advise. That’s a meaningful step, and it’s being taken carefully, with a small test rather than a global switch-flip.
For readers here, the useful takeaway isn’t “Google added shopping.” It’s that the assistants you’re already talking to are slowly being given hands. Retail happens to be the first place those hands show up, because the task is well-defined and the payoff is obvious.
Watch what happens in October. A broader rollout would tell us this cleared its internal bar. A quiet one would tell us something too.
🕒 Published: