The shareholder letter can be accessed here.

“AI is our friend, not a threat,” Eternal (formerly Zomato) co-founder Deepinder Goyal wrote in the company’s Q4FY26 letter released April 28. Eternal has expanded its OpenAI partnership, built a system allowing AI to place food orders on a user’s behalf, and joined a live agentic payments platform doing exactly that. Indian law has no framework for when it goes wrong.

Goyal draws on Google’s history to make his case. Despite Google’s scale and distribution advantage in pulling transactional behaviour into Google Flights, Hotels, Shopping, and restaurant ordering on Maps, vertical apps retained their users. “And yet Booking.com is still here. Expedia is still here. Amazon is still here,” he wrote, arguing that general-purpose interfaces are poor substitutes for apps consumers have built habits around.

He does not claim the threat is impossible. “If something similar didn’t work in the past, it doesn’t mean it won’t work in the future,” he wrote, before concluding: “At this point, we believe there is nothing to panic about.” He added the company would keep its eyes open “without an iota of complacency or wishful thinking.”

He makes three arguments for why Eternal’s platforms are resilient:

• Physical infrastructure: “Most of our business happens in the offline realm. AI will not replace the need to move food through traffic. It will not stock shelves in a dark store. It will not manage crowd flow at a 50,000-person event. The physical world doesn’t get disrupted by language models”
• Consumer habit: People who order on Zomato four times a week or on Blinkit every other day “are not going to reroute those habits through a chat window,” with AI assistant users described as “a real but early segment, and currently, nowhere close to disrupting our business”
• Brand recall: Each of Eternal’s apps owns a specific trigger in the customer’s mind and that mental real estate “is not something that gets dislodged easily”

Eternal claims to already be one of India’s largest AI deployers. Goyal claims Eternal is “one of the largest deployers of AI in India,” with systems already live across demand prediction, route optimisation, supply chain management, fraud detection, and partner support. “These systems are trained on billions of real interactions, where the cost of getting something wrong is immediate and measurable,” he wrote, drawing a distinction between Eternal’s operational AI and general-purpose chat interfaces.

The bigger argument: does AI expand the market or expose it? “AI lets us serve people and businesses that were previously too hard, too small, too fragmented, or too operationally expensive to serve well,” Goyal wrote. Conversational interfaces, he argues, let customers “express intent in full sentences rather than two-word search queries, which is a far richer signal to convert against,” meaning a stronger basis for matching users to relevant results and driving purchases. What Eternal has built toward this includes:

• Expanded its OpenAI partnership in February 2026 to deploy AI tools across Zomato, Blinkit, District and Hyperpure, covering internal workflows, partner-facing systems, and user-facing search and discovery
• Built a Model Context Protocol (MCP)-based system on Zomato that allows food ordering through text prompts. Zepto and Zomato have both tested similar systems publicly while keeping them out of consumer-facing products
• Launched Healthy Mode and Natural Language Search (NLS) on Zomato to reduce friction for older users, non-native language speakers, first-time online buyers, and tier 2 and tier 3 city consumers
• Using AI for delivery partner onboarding to reach people who struggle with documentation or language
• Deploying AI-based growth guidance for restaurant partners from day one

Agentic commerce on Zomato is already live. Razorpay and the National Payments Corporation of India (NPCI) announced agentic payments on Claude in February 2026, allowing users to order food from Zomato, Swiggy, and Zepto without leaving a conversation. Under the system, users set a one-time spending limit upfront and an AI agent then completes transactions on their behalf within those limits using Unified Payments Interface (UPI) flows. Zomato is already one of the platforms on which this system operates.

Indian consumer protection law has no answer for what happens when this goes wrong. Three frameworks are relevant and all fall short for specific reasons:

• The Consumer Protection Act, 2019 and its e-commerce rules require platforms to display grievance mechanisms and honour refunds, but assume the buyer made a conscious purchase decision. If an AI agent made that decision, it is unclear who files the complaint and whether the platform bears liability.
• The Digital Personal Data Protection (DPDP) Act requires consent for data processing, defining it as a specific, informed, voluntary act by the data principal. A blanket authorisation to an AI agent to place orders does not clearly satisfy that definition.
• Section 79 of the Information Technology (IT) Act protects platforms from liability for third-party content when they act as neutral conduits. An AI agent placing orders on a user’s behalf is a platform-initiated action, not third-party content, which puts it outside the safe harbour entirely.

The users Goyal names are the least protected. Goyal argues conversational interfaces lower friction “especially for older users, non-native language speakers, first-time online buyers, and tier 2 and 3 city consumers who may struggle with app flows.” These are also the users least likely to understand they have delegated a transaction to an AI system, least likely to read consent flows carefully, and least equipped to navigate a dispute when something goes wrong. Goyal’s market expansion argument and his dismissal of regulatory risk point in opposite directions: the bigger the expansion into these user segments, the more acute the protection gap becomes.

Platforms are moving faster than the law. When agents initiate transactions autonomously, responsibility splinters across developers, deployers, cloud providers, and integrators, with no framework for cascading liability, audit trails, or clear attribution. MediaNama has reported on this governance gap in detail.

MediaNama tested Swiggy’s MCP integration when it was released and found platforms drawing the line cautiously, with gated access, confirmation steps, and cash-on-delivery-only payments indicating where companies are limiting AI autonomy for now. The caution is commercial, not regulatory.

At the India AI Impact Summit 2026, experts called for three infrastructure pieces to make agentic commerce trustworthy: an Aadhaar equivalent as a foundational identity layer for AI agents, a UPI equivalent for how money flows between agents, and an Open Network for Digital Commerce (ONDC) equivalent as an interoperable platform. None of these exist yet, and neither MeitY nor the Ministry of Consumer Affairs has issued any guidance on agentic commerce liability or consent frameworks.

Goyal wrote that “at this point, we believe there is nothing to panic about.” The users he wants AI to bring onto Eternal’s platforms may not have the same luxury.

Also read:

• Zepto Cafe Tests AI Agent That Can Place Orders For Users
• Ordering Via ChatGPT on Swiggy Services Isn’t Quite Working
• Eternal Expands Deal With OpenAI: What This Means
• India Faces AI Accountability Crisis as Autonomous Agents Rise