Suppose someone hands you the keys to a Lamborghini Huracán. You live in Delhi 6, a short walk from Karim’s, where the kebabs arrive hot every evening.
So, our stomach is looked after.
Now the car.
Brian Chesky told TechCrunch’s Ivan Mehta on Oct. 1 that AI agents need an operating system of their own.
Coming back to the car, Huracán reaches 100 km/h in about three seconds.
Chandni Chowk lets it reach eight, behind a rickshaw,and the fuel gauge drops at a couple of kilometres to the litre. That’s where many AI agent plans sit today and Chesky explained why in a TechCrunch interview on Oct. 1.
The Compliant
Chesky said today’s AI apps run on operating systems built for people tapping screens and he says nobody is building the full operating system and developer kit that agents would need.
He wants AI working at the kernel level, with agents talking to each other, instead of a race to become the one main agent that handles everything. (Muse?? Anyone?)
The Potholes
Agents read price, terms, ratings and availability.
They ignore layout, brand colours and countdown timers. If they handle a big share of daily purchases, an app risks becoming a catalogue that competes on whatever a machine can sort. Apps have also worked together only when two companies signed a deal. And chatbots send leads to apps, which puts them upstream of the sale. Chesky tried Airbnb through Muse and Instinct and said it worked poorly.
The Open Lanes
He expects apps to stay central while Apple and Google control the main platforms. He pointed to travel studies saying people enjoy planning a trip more than taking it and argued that a designer can prompt better software than a consumer can. Airbnb plans voice agents this fall and multiplayer AI for groups within three to six months.
I do think an agentic operating system would settle identity and permissions once and in my reading it would also hold the audit trail.
That matters to anyone who has to explain an automated decision. But I’d like apps to stay too. An app is where a company explains its terms in its own words and earns a second visit. Chesky has an obvious stake here and shoppers may gain from strict comparison.
All that said, the risk to app businesses is still real.
Five dos and don’ts for app-dependent businesses
Do publish prices, fees and cancellation terms through APIs and MCP, as current as your screens.
Don’t leave the main facts in PDFs, and don’t let the API drift from the app. An agent that finds your data disagreeing with itself will pick a rival whose numbers match.
Design the handoff. Chesky named what people still want an app for: browsing, comparing, messaging a host, verifying identity and using a map.
Don’t block agent traffic. A wall sends the customer to whoever lets the agent in.
Do compete on facts a buyer can verify: price, delivery record, reviews and return terms.
Don’t lean on countdown timers or pre-ticked add-ons. An agent has no pulse to quicken.
Do protect the direct relationship through identity, loyalty, service after the sale and data you hold with consent.
Don’t let one agent own your customer. A lead source today can become your landlord tomorrow.
Do fix your own processes so your agents can finish a task start to end. Start with one workflow, refunds say.
Don’t bolt a chatbot onto a legacy workflow and call it agentic, and don’t wait for Apple or Google to build the new platform.
The Lamborghini stays in the garage until the road exists. I love Delhi 6, and I’m still walking to Karim’s.