Indian quick commerce platforms are cautiously stepping into a new phase of artificial intelligence — one where systems do not just assist users but act for them. Grocery and quick commerce platform Zepto has built an internal AI tool that allows a large language model (LLM) to place Zepto Cafe orders using natural-language instructions, marking an early experiment in what is known as agentic commerce.
From assistive AI to delegated action
Unlike chatbots that recommend items or answer queries, Zepto’s system can complete an entire transaction — selecting items, choosing an address and proceeding to checkout. This shift raises deeper questions around consent, accountability and liability, especially when AI actions involve money, inventory and delivery.
Significantly, Zepto has kept the tool away from its consumer-facing app and website. The internal-only approach signals that once AI begins acting autonomously, errors are no longer harmless suggestions but completed purchases with real-world consequences.
How the system works
The experiment uses the Model Context Protocol (MCP), an open-source standard developed by Anthropic, to structure decision-making. MCP translates user instructions into a sequence of actions, which are then executed by Playwright — a browser automation tool that mimics human interactions such as clicking and typing.
By operating entirely through a regular web browser, the AI behaves like a delegated user rather than a privileged system. It does not access Zepto’s internal APIs or backend infrastructure, allowing the company to test agentic ordering without redesigning core systems.
Why Zepto is keeping it internal
The tool was built by a Zepto engineer and shared on GitHub before being amplified by Co-Founder and CEO Aadit Palicha on LinkedIn. However, using it requires manual setup and browser login, reinforcing that it remains an experiment rather than a product feature.
Once AI places orders independently, disputes over unintended purchases, refunds and responsibility become complex. Industry leaders increasingly argue that AI should act as a co-pilot, with humans retaining final control over high-impact decisions like payments.
A broader industry trend
Zepto is not alone. Zomato has tested similar AI-driven ordering layers, while payments firms including Razorpay and NPCI, alongside OpenAI, are piloting agentic payments via ChatGPT, with BigBasket among early participants.
For now, these experiments remain tightly controlled. The message from Indian consumer platforms is clear: AI-driven autonomy may be powerful, but it must be tested quietly — and carefully — before being trusted at scale.