Brian Chesky Is Underestimating Chatbots

An advisor recently sent me this clip of Airbnb CEO Brian Chesky talking about AI, travel, and e-commerce.

His argument was provocative: chatbots are not the right interface for categories like travel and e-commerce.

He points to four limitations:

• Shopping is visual, while chat is primarily text.
• Chat lacks the direct manipulation of filters and controls.
• It makes comparison harder.
• It is mostly single-player, while many purchases are collaborative.

I understand the critique. But I think these are mostly limitations of today’s chatbot interfaces, not fundamental limitations of AI-driven commerce.

I recently used Instinct, a primarily text-based AI platform, to plan a Thanksgiving trip for my family and flights for an upcoming conference. It already worked remarkably well.

The experience can absolutely get better. More visual. More interactive. More collaborative.

But that is the point.

The future is probably not a text chatbot replacing a search-results page. It is an AI agent that understands what you are trying to accomplish and uses the right interface for the task.

Brian’s point: shopping is visual, while chat is text-based

I agree with the first half of that.

Shopping is often visual. But AI does not need to be text-only just because the interaction starts with text or voice.

We are already seeing AI products incorporate photos, maps, tables, comparisons, suggested actions, and interactive interfaces directly into their responses.

Traditional marketplaces generally build one interface that has to work for every shopper: the same filters, cards, maps, and search-results pages.

AI can be more flexible.

If the task calls for photos, show photos. If it calls for a map, show a map. If it calls for a comparison table, build one.

The interface can adapt to the shopper’s intent instead of forcing the shopper into a predefined interface.

So I do not think the real question is whether shopping is visual. It is whether AI can present information visually when that is the best way to help the shopper.

I think the answer is clearly yes.

Brian’s point: chat lacks direct manipulation and filters

This is probably the critique I understand most.

Sometimes clicking is simply faster. If I know I want a hotel under $500 a night with a pool, a price slider and checkbox work pretty well.

AI should not eliminate those interactions just because conversational interfaces are fashionable.

But filters also have a limitation: they force consumers to express what they want using categories someone else decided in advance.

That works well for price, dates, and number of bedrooms. It works less well when the decision involves a combination of preferences, priorities, and tradeoffs.

An agent can start with what the consumer actually wants, translate that into relevant search criteria, and then surface filters or suggested refinements when they are useful.

So I do not think this has to be conversation versus filters.

Traditional search asks you to translate your intent into the marketplace’s taxonomy. AI can translate your intent into the search.

Brian’s point: AI makes comparison harder

This is where I disagree most strongly.

Yes, people want to compare. But I am not convinced the best way to help someone compare is to give them hundreds of results and ask them to narrow the list themselves.

When I was building Autolist, one of my early mentors told me that the perfect car search engine would give you one car: the car you should buy.

At the time, I disagreed. I thought people needed to understand why, see the alternatives, and evaluate the tradeoffs.

Years later, I think we were both partly right.

The ideal experience is probably not 100 results. It also is not one unexplained recommendation.

It is a recommendation, the reasons behind it, and the best alternatives.

Instead of returning hundreds of cars, an agent can understand what you are optimizing for, narrow the market, recommend the best fit, and explain why several other options might be better depending on the tradeoffs you care about.

That is comparison too. In many cases, I think it is a better version of it.

Brian’s point: chat is single-player, while shopping is collaborative

Brian is right about the underlying behavior.

Travel, cars, homes, and other considered purchases are often multiplayer decisions.

But I am not sure today’s marketplaces have solved this particularly well either.

AI conversations are already easy to share, and it is not difficult to imagine shared sessions where multiple people contribute preferences, react to options, and move toward a decision together.

The current experience is imperfect.

But I see that as a product limitation to solve, not a reason agents are fundamentally wrong for these categories.

One other advantage Brian’s framework does not really address

There is another difference that may matter even more than the interface.

A marketplace can help you search the marketplace. An agent can potentially help you search the market.

If I am planning a vacation, my real question is probably not, “Which Airbnb should I book?”

It is, “What is the best trip for my family?”

The answer could be an Airbnb. It could be a Vrbo. It could be a hotel where I have loyalty status or points. It could be a direct booking with a better offer.

The marketplace starts with its supply.

The agent can start with my objective.

That is a much more interesting shift than replacing a search box with a chat box.

There is a larger conversation to have about what happens to marketplaces as agents become more important, including whether marketplaces themselves become more like data and transaction layers for AI. We will come back to that in a future piece.

Where I think Brian and I agree

Toward the end of his comments, Brian makes an important distinction: he believes the future is agentic, even if he does not believe today’s chatbot interface is the answer.

On that point, I think we largely agree.

Maybe the mistake is treating the chatbot as the product.

Chat is one interface into an agent.

Sometimes the right experience will be conversational. Sometimes it will be visual. Sometimes it will look like traditional search. Sometimes the agent may simply recommend the best option and explain the alternatives.

The important part is the intelligence underneath: understanding what the consumer is trying to accomplish and helping them get there.

Why this matters for automotive

Automotive has nearly all of the characteristics Brian is talking about.

It is visual. It is comparative. The decision involves a lot of tradeoffs. It is often collaborative.

That is exactly why I think agents will be powerful here.

At CarGenius AI, we are not trying to build the consumer’s agent. We are building the infrastructure that lets AI systems access dealer data and capabilities directly.

The future of commerce may very well not be a chatbot.

But I increasingly believe it will be agentic.