#e4mXplains:  Why India is a stress test for agentic commerce

Imagine, dear reader, that it is 11pm and you have run out of ice.

There are, increasingly, two visions of how technology might solve this problem. In the first, you tell an AI assistant that you need ice. The agent understands the request, finds a merchant, navigates the transaction and — in the increasingly agentic future being built by Google, OpenAI and others — perhaps even checks out on your behalf.

Or, if you live in an Indian metro, you open Blinkit, Zepto or Swiggy Instamart, tap twice, pay and go back to whatever you were doing. The ice may arrive before you have finished explaining agentic commerce to somebody.

And therein lies a peculiar problem for one of AI’s most heavily advertised next acts.

Agentic commerce is supposed to remove friction between wanting something and buying it. But India has spent the last several years turning the removal of friction into an industry of its own. For millions of urban consumers, particularly when it comes to food, groceries and everyday essentials, the journey from impulse to transaction has already been compressed almost to its physical limit.

So what, exactly, is the agent fixing?

That question matters because agentic commerce is rapidly moving beyond demos. OpenAI launched Instant Checkout in ChatGPT last year, initially allowing US users to purchase eligible Etsy products directly in chat, while its Agentic Commerce Protocol is designed to let merchants plug their catalogues and checkout systems into AI interactions. 

Google is building its own version of the stack. Its Universal Commerce Protocol, launched this year, provides a common language through which AI agents, merchants and payment providers can handle product discovery, carts, checkout, orders and loyalty. Google describes UCP as part of the “foundation for agentic commerce” in India, even as its current UCP-powered checkout eligibility remains limited to select markets outside India. 

The premise is compelling. Instead of discovering a product in one place, comparing it somewhere else, opening a merchant app, entering payment information and checking out, an agent could increasingly collapse those steps into a conversation.

But that proposition assumes those steps remain meaningful sources of friction.

India already removed some of them

India’s quick-commerce economy is an awkward counterexample.

Redseer estimates quick commerce has already become a $10-billion-plus GMV market with more than 30 million monthly users, with metros accounting for more than 80% of business. Another 2026 Redseer analysis estimates the channel now represents roughly 17% of Indian online retail and was growing at around 120% year-on-year in FY26. 

More importantly, quick commerce has expanded far beyond emergency milk and forgotten onions. Beauty, personal care, electronics, toys and other categories are increasingly being absorbed into the same behaviour. Redseer now describes quick commerce as a structural growth engine rather than a temporary pandemic-era convenience. 

Payments have undergone a similar compression. UPI processed nearly 25 billion transactions in August alone, according to NPCI data cited by Reuters, with more than 550 million users. 

In other words, the Indian consumer does not necessarily need an AI agent to eliminate the tedious steps between “I need toothpaste” and “toothpaste purchased”.

Those steps have already largely disappeared

And in food and quick commerce, the difficult part of the transaction is not necessarily discovering the product or paying for it. It is getting a physical object from a restaurant, dark store or warehouse to somebody’s door in minutes.

That distinction matters enormously.

AI can remove digital friction. It cannot wish away logistics

Agentic commerce works particularly elegantly when the underlying fulfilment system already exists independently of the interface.

An AI can compare flights, find an itinerary and facilitate a booking; the airline still flies the aircraft. It can compare hotels; the hotel still provides the room. It can help someone choose between laptops, evaluate specifications and eventually hand the purchase to a retailer that already possesses warehousing and delivery infrastructure.

Those are high-consideration purchases where the cognitive work (comparison, search, filtering and decision-making) can itself be substantial.

Food and quick commerce invert that equation.

Here, the decision may take seconds while the expensive engineering sits behind the screen: dark stores, restaurant integrations, inventory management, riders, routing algorithms and last-mile operations.

Google’s UCP documentation reflects that reality. Merchants remain merchants of record, while fulfillment remains part of the underlying commerce system. UCP standardises the interaction between agent and merchant; it does not create a fleet of delivery riders. 

Which raises the harder question: if Swiggy, Zomato, Blinkit or Zepto already own both the consumer habit and the infrastructure required to fulfil it, why should they surrender the interface to an outside AI agent?

They may instead simply make their own interfaces agentic.

The agent may have to move up the value chain

None of this means agentic commerce has no future in India.

It may mean the opposite: India could reveal where it is actually useful.

For a ₹300 grocery basket, additional reasoning may have little value. For a ₹70,000 television, an international holiday, a complicated insurance product or a purchase involving financing and trade-offs across several merchants, an intelligent agent capable of interrogating choices could remove genuine friction.

OpenAI itself frames its shopping proposition around precisely this problem: product discovery when the consumer is still deciding, comparing and trying to make sense of multiple options. 

That suggests agentic commerce may be most valuable not where transactions are frequent and simple, but where they are infrequent and cognitively expensive.

Which makes India an unusually revealing market.

The country is already one of Google’s largest commerce and discovery environments: Google says Search and YouTube feature in 93% of Indian consumer journeys involving discovery of a new brand, product or retailer, based on a Google-commissioned Ipsos survey. Pine Labs and Google Cloud are meanwhile explicitly working to prepare Indian merchants for agentic commerce. 

But the country has also built a convenience economy that is remarkably good at solving certain purchases without an agent.

That makes India less a laboratory in which agentic commerce simply gets deployed than a market in which its assumptions get tested.

AI’s promise is to remove friction.

In India, in some of the most common digital purchases, the more uncomfortable possibility is that the agent itself becomes the extra step.