Shopping without searching: The Agentic AI future Amazon, Swiggy and Zepto are building in India

There is a number worth sitting with before the Prime Day deal alerts start firing on July 4. During the 2025 holiday shopping season in the US, roughly one in five online orders, about 20%, was influenced by an AI agent or shopping assistant, according to Salesforce. Adobe, tracking the same window, found that traffic arriving at US retail sites from generative AI tools had jumped nearly 700% year over year. A behaviour that barely registered eighteen months ago is now a measurable slice of how the world shops.

The money is following the behaviour. Grand View Research pegs the global agentic commerce market at about $5.7 billion in 2025, rising to $7.7 billion in 2026 and $65.5 billion by 2033. Bain's wider definition, which counts agent-influenced purchases, puts the US alone at $300 to $500 billion by 2030, as much as a quarter of all e-commerce. Gartner expects a fifth of digital commerce transactions to run through AI platforms by the end of the decade. The estimates disagree wildly, mostly because everyone is measuring a slightly different thing, but the direction is not in dispute.

The shift is being driven by what's known as agentic AI, systems that go beyond generating content to actually completing tasks. Rather than simply responding to prompts, these AI agents can understand intent, reason through multiple steps, access the tools they need and take action, whether that's recommending a product, processing a refund or placing an order.


The future is arriving faster than many expected. At Google I/O this year, Google unveiled shopping-focused Gemini capabilities that can discover products, compare options, track prices and even complete purchases on a user's behalf.

Alongside it, the company introduced the Universal Commerce Protocol (UCP), an open standard designed to let AI agents communicate directly with retailers and commerce platforms. Built to work alongside emerging standards such as Model Context Protocol (MCP) and Agent2Agent (A2A), the initiative has attracted support from major retailers and payment companies, signaling that the industry is beginning to converge around a common language for AI-powered shopping rather than isolated ecosystems.

For India, the shift arrives with its own accent. Mordor Intelligence estimates that nearly half of Indian retailers, 48%, are already piloting generative AI, and it expects Asia-Pacific to be the fastest-growing region in the world for retail AI, compounding at roughly 35% a year through 2031. That is the backdrop against which three of the platforms Indians use most, Amazon, Swiggy and Zepto, are each making a distinct bet on what "agentic" commerce actually means. We enquired with all three to understand how they are using agentic AI, and the results are interesting. Especially because Amazon plays in everything e-commerce, while Swiggy and Zepto are in quick commerce-centric.

Amazon: teaching the store to remember you

For Kishore Thota, Director of Shopping for India and Emerging Markets at Amazon, the through-line across a decade of experiments has been persistence rather than any single technology.

"We stay committed to an idea until it succeeds," he says, pointing to Amazon Live as an example of a product that took years to find its audience. Generative AI, by contrast, has delivered immediate customer value.
At the centre of Amazon's strategy is a customer memory system that looks beyond purchase history. Instead, it attempts to understand whether a shopper is value-conscious or brand-conscious, frequently travels, shops for a family or has particular hobbies. That context is combined with large language models to personalise recommendations.

Amazon also lets customers ask a simple question: "What do you know about me?" The system then explains, in plain language, the preferences it has inferred.

That memory powers features such as personalised recommendations, Quick Summary, which answers whether a product is right for a customer, and price transparency tools that show 30-day and 90-day price history before a purchase.

Thota argues that the objective is not to persuade customers to buy, but to give them more context before making a decision.

The company is also expanding Rufus, its AI shopping assistant, alongside Prime Day, which returns to India from July 4 to 6.

Zepto: AI agents for the moment it all goes wrong

Zepto is applying agentic AI to one of the biggest pain points in quick commerce: customer support. Instead of relying on a single chatbot, the company has built an in-house system called Zap that uses multiple specialised AI agents, each responsible for a specific task. When a customer raises a complaint, such as a missing item, damaged product or refund request, the system first authenticates the user and gathers order context before routing the issue to a dedicated AI agent trained for that category.

Before making a decision, Zap runs multiple verification checks. For image-based complaints, separate AI agents validate whether the uploaded photo matches the reported issue, confirm it belongs to the correct order and even detect if an image has been reused or manipulated. Another specialised agent evaluates the product itself, checking for spoilage, damage or expiry. Only after these checks does a policy engine determine the appropriate resolution, whether that's an instant refund, replacement or another action. Human support agents are brought in only when a case is too ambiguous or requires manual intervention.

During ET's visit to Zepto's Mother Hub in Bengaluru, we got a first-hand look at how the company has built much of this ecosystem itself. Zepto isn't just developing its software stack in-house, but also the hardware powering its fulfilment network, allowing the two systems to work together seamlessly.

According to the company, Zap has reduced average support resolution times by 75%, from minutes to seconds, increased positive customer reviews by 20%, and maintained consistent performance during peak demand. You can also watch our behind-the-scenes video from the Mother Hub to see how Zepto's custom-built hardware and software work together.

Swiggy: betting that the interface itself disappears

Swiggy is making perhaps the boldest bet of the three. Rather than simply improving its own app, it is preparing for a future where the app itself becomes optional.

"For over a decade, we've built the technology and fulfilment network that powers on-demand commerce at scale," says Vivek Garg, Vice President of Engineering at Swiggy. "Today, AI is helping us personalise experiences, resolve support queries faster, optimise delivery routes, forecast inventory and accelerate refund resolution."
The bigger shift comes through Swiggy's MCP rollout and Builders Club initiative, which opens its commerce infrastructure to external developers and AI agents.

Built on AWS, the platform exposes multiple MCP servers and API tools across Food, Instamart and Dineout, allowing approved AI assistants to order food, buy groceries or reserve restaurant tables without the user opening the Swiggy app.

Garg believes commerce will increasingly be driven by intent rather than interfaces.

"A consumer could ask an AI assistant to order dinner, schedule a grocery delivery and reserve a table for the weekend, and the entire workflow can happen seamlessly in the background."

If consumers increasingly begin shopping inside AI assistants rather than apps, Swiggy believes the winning platforms will be those that AI agents can reliably understand and transact with.

The same destination, three different roads

Viewed together, the three strategies reveal where commerce is heading.
Amazon is using AI agents to make discovery smarter and recommendations more personal, reducing the effort required to find the right product.

Zepto is applying AI after the purchase, turning what was once a frustrating customer support experience into an automated workflow that resolves routine issues in seconds.

Swiggy is making perhaps the biggest structural bet of all: that in the future, consumers may never need to open its app, with AI assistants placing orders directly wherever user intent originates.

Different strategies, but the same destination.

The broader industry is moving in the same direction. Google wants Gemini to become a shopping agent that can browse, compare and buy across the web. Swiggy is exposing its commerce stack through MCP so outside AI assistants can transact directly. Amazon is teaching its store to remember customers and make increasingly personalised recommendations. Zepto is automating everything that happens after the purchase. Together, they point to a future where AI is no longer just answering questions, but increasingly acting on our behalf.
Prime Day may begin on July 4, but it also marks something larger. A shopper asking Rufus for advice, a Zap agent resolving a refund before any human intervenes, or a Swiggy order placed entirely through an AI assistant are no longer demonstrations of what's possible.

They are early signs of a future where shopping starts with intent, not search, and where you may not even have to decide what to buy.