Nilay Patel’s recent article in The Verge explores “the DoorDash problem” - what happens when AI agents sit between service providers and customers. It surfaces a critical question: which platforms will survive the AI disruption?
Amazon just sued Perplexity to stop its AI browser from shopping on Amazon.com. This isn’t just legal posturing; it’s a signal of how platform companies will defend their business models as AI reshapes customer relationships.
What’s at stake: Companies like Amazon, DoorDash, and Uber built massive businesses not just on the transaction itself, but on everything around it - ads, subscriptions, impulse purchases, brand placements. If AI agents reduce these platforms to invisible databases competing purely on price, that ancillary revenue disappears and the economics collapse.
Service providers are splitting into two camps:
- The Confident (Uber, TaskRabbit): They welcome AI agents. Their moat is operational complexity—supply networks, logistics, and verification.
- The Panicked (Amazon): Amazon sued immediately. As Nilay points out, “Amazon’s already turned itself into commodity provider of weirdo drop-shipped products from alphabet soup companies.” Their $56B+ ad business depends on owning the customer interface.
Are these reactions justified?
If we judge by current capabilities, the threat feels overblown. Today’s shopping agents are essentially glorified search bars. These models often fail to grasp nuance, struggle with complex user intent, and can’t reliably navigate the web. But looking forward, the panic makes sense.
AI agents are already moving toward context enrichment - understanding you across conversations, building comprehensive user profiles, learning from chat history. When you type “buy dumbbells”, your agent will already know that your doctor mentioned bone density concerns, you have limited space, and that past gym memberships failed.
Assuming the reliability problem is solved, the agent will make the best shopping decision for you based on all internal and external factors. That decision might be: start with resistance bands, and don’t get them from a third party reseller on Amazon but directly from a reputable manufacturer’s site.
This creates a natural fit for platforms that profit mostly from execution excellence. Uber (logistics), TaskRabbit (vetted labor), and ZocDoc (verified providers) remain valuable because the AI needs them to actually do the work.
But companies relying on interface control - ads, impulse buys, and subscription lock-in - face a different future. When agents understand what users actually need, the “browse and buy” mechanics break down.
I’m curious how this plays out in PropTech. Zillow and Realtor.com face similar vulnerabilities to Amazon (ad-driven, lead gen). How will the industry change when agentic AI finally arrives?
Still thinking through the implications.