Brian Chesky says AI agents need an operating system, not just a chatbot

Airbnb is pushing its consumer AI strategy beyond a chatbot, but CEO Brian Chesky says the larger industry still lacks the software infrastructure needed for agents to work reliably across apps.

In an interview with TechCrunch AI, Chesky said Airbnb wants to make its platform more compatible with AI agents while preserving browsing, comparison, messaging, maps, identity checks, and other functions that make travel planning a visual and collaborative activity. His argument is that the next stage of consumer AI will require more than a single assistant that accepts text commands: it will need an operating system, software development tools, and common standards that let agents and applications work together.

The comments arrived as Airbnb rolled out a new AI-powered search experience in its fall product update. They also offer a view into how a major consumer platform is approaching agents at a time when startups are trying to make software act on behalf of users.

Airbnb wants agents without losing the travel experience

Chesky has long argued that a conventional chatbot is poorly suited to travel discovery and e-commerce. In the interview, he repeated that view, saying chat interfaces tend to present only a small number of options and require several turns before a user reaches a useful result.

That interaction model may work for a direct request such as booking a flight, he said, but it can remove the inspiration and browsing that many people value when planning a trip. Airbnb’s current search interface is therefore not being presented as a final design. Chesky expects the company to develop an intermediate model that combines conversational assistance with richer visual controls.

He also described travel planning as a social activity. Over the next three to six months, Airbnb plans to explore what Chesky called “multiplayer” AI: interfaces that allow several people, such as family members or friends, to participate in the same planning process.

This is a significant constraint for AI product teams. An assistant designed around one user and one conversation may be efficient for individual tasks, but it is less suited to decisions involving preferences, negotiation, and shared discovery.

Chesky’s operating-system thesis

Chesky’s broader criticism is aimed at the current race to become the primary AI agent through which users access everything. He said companies are building agents on top of iOS, macOS, and Windows without creating an operating system designed around AI in the first place.

In his proposed model, agents would operate closer to the core of the platform, with applications exposing capabilities through a software development kit. Those agents could then exchange information and hand off tasks without forcing users to navigate disconnected interfaces.

He pointed to ChatGPT’s earlier attempt at an app store as an example of the problem. In Chesky’s telling, an app marketplace cannot function like Apple’s App Store without both an operating system and a developer framework that allow applications to provide robust functionality. He said Airbnb’s own experience using consumer agents such as Muse and Instinct has not yet produced reliable results, particularly for booking accommodations.

Chesky also sees interoperability as a potential role for the Model Context Protocol, or MCP, which he described as a standard that can connect agents to software. His longer-term concept is for Airbnb itself to expose multiple specialized agents, potentially coordinated by a broader Airbnb agent and connected to agents from other services.

Evidence and limits behind the vision

The confirmed product development in the interview is Airbnb’s new AI-powered search, its stated effort to become more agent-friendly, and its plans to investigate collaborative and voice-based experiences. Chesky also said Airbnb expects to introduce voice agents for search and customer service in the fall.

The larger operating-system proposal is an executive thesis, not a demonstrated industry architecture. The interview provides no independent benchmark showing that Airbnb’s AI search improves conversion, reduces support workload, or performs better than competing travel tools. Chesky’s assessment that agents currently handle hotels and Airbnb listings poorly is also an executive observation rather than a controlled evaluation.

Likewise, his claim that AI is helping Airbnb ship features faster and reducing the number of meetings he needs is based on his own account. It should not be treated as evidence of company-wide productivity gains without additional measurement.

The limits matter because agent performance depends on more than access to data. Booking travel requires authentication, payment, availability checks, policy interpretation, user confirmation, and the ability to recover when information changes. A basic connection between an agent and an application does not automatically solve those reliability and accountability problems.

What the model means for builders and enterprises

For product teams, Airbnb’s position suggests that adding a chat box to an existing application is unlikely to be enough. Companies preparing for agents may need to expose structured actions, permissions, status information, and clear handoff points while retaining interfaces that users trust for exploration and verification.

Travel is an especially demanding test case, but the same issues apply to commerce, financial services, healthcare, and enterprise software. An agent may identify an option, while the application still needs to display the details, confirm the user’s intent, enforce access controls, and record what happened.

The competitive question is also changing. If agents become important traffic sources, platforms may gain new leads from assistants while risking a weaker direct relationship with customers. Chesky said chatbots have so far appeared useful for lead generation, but Airbnb’s effort to remain “agentic” indicates that the company does not want to surrender the rest of the customer journey.

For enterprise buyers, interoperability should therefore be evaluated alongside automation. The relevant questions include whether an agent can move safely between systems, whether users can inspect and override actions, and whether the underlying application remains usable when automation fails.

What to watch next

The first signal will be Airbnb’s announced voice-agent rollout for search and customer service, including how much of the interaction can be completed without a human handoff. Its performance will matter more than broad predictions about voice replacing typing.

The company’s next three to six months of work on multiplayer AI will show whether shared planning can move beyond a demo into a dependable product workflow. Developers should also watch for concrete Airbnb APIs, SDK features, permission models, or MCP integrations rather than relying on general statements about agent readiness.

Across the market, the key test is whether Apple, Google, or another platform provider offers a genuine agent framework with common controls and interoperability. Without that layer, Chesky expects apps to keep adding agents inside their existing products instead of becoming parts of a coherent agent ecosystem.

Creati.ai perspective

Chesky’s argument is useful because it shifts attention from the assistant’s conversation quality to the application infrastructure underneath it. Consumer agents will not become dependable simply by generating better responses; they must be able to perform bounded actions, use rich interfaces, coordinate with other people, and explain when a task requires a human or a conventional app screen.

Airbnb’s strategy also highlights a tension that product leaders should not ignore. Automation can shorten a transaction, but travel discovery depends on browsing, comparison, and shared judgment. The strongest agent products may therefore be hybrids: capable of acting quickly when the user knows what they want, but able to preserve control and context when the decision itself is part of the experience.