Zepto

The demand aggregation and catalog management layers are vulnerable — AI can predict what you need before you know it and automate procurement from suppliers. But someone still has to physically move the mangoes from warehouse to doorstep, and that last-mile choreography in Indian cities remains stubbornly analog for now.

Delivery riders will be the last to go — navigating Bangalore's Silk Board with a bag of eggs requires human-level spatial reasoning and traffic negotiation skills that robotics hasn't cracked. Warehouse pickers and inventory managers, however, are looking at computer vision and robotic fulfillment systems that don't take chai breaks.

This is one of the few genuinely physical moats left in Indian tech: 300+ dark stores positioned within 2km of high-density demand, each stocked based on hyperlocal purchase data that took years to accumulate. Replicating that footprint requires hundreds of crores and 18 months minimum — assuming you can even get the real estate.

They're hiring for 'AI/ML engineers' and 'data scientists' but the job descriptions read like demand forecasting and route optimization — useful but incremental. The real question is whether they're preparing for autonomous delivery or just hoping their warehouse density buys them enough time to figure it out.

People will always need groceries, and they'll always want them fast — AI doesn't eliminate hunger or impatience. The need is rock-solid; the question is whether the delivery mechanism evolves from riders to robots, and whether Zepto's warehouse network becomes an asset or a liability in that transition.