The Limits of Chatbots
Airbnb rolled out its new AI-powered search as part of its fall update, a significant step for a company that has taken a measured approach to artificial intelligence. Co-founder and CEO Brian Chesky has maintained that a chatbot-only interface is unsuited for travel discovery, prompting a discussion on interface evolution, consumer AI limitations, and the role of voice technology.
Chesky's stance on chatbots remains unchanged. He notes they are suboptimal for browsing or shopping because they present limited options at a time, requiring multiple turns to reach a result. However, he views Airbnb's current search iteration as an intermediate step rather than a final destination, expecting the interface to evolve.
Chesky noted that a chatbot is not the right interface for e-commerce, and today's implementations are not the endgame for travel or shopping. He envisions an interface positioned between a traditional chatbot and early app versions.
While some interactions benefit from speed and automation, travel planning relies heavily on inspiration and visual browsing. Studies indicate that people often derive more pleasure from anticipating and planning travel than from the trip itself, making it crucial to preserve the library-like experience of dreaming and exploring.
Collaborative and Generative Interfaces
Another limitation of standard chatbots is their single-user design. Because travel planning is frequently collaborative, involving family or friends, Airbnb is exploring "multiplayer" AI tools over the coming months to enable multiple users to interact simultaneously.
Industry-wide experimentation is pushing beyond static design. Startups are testing "generative" interfaces—screens built dynamically by AI rather than pre-designed by developers. Chesky expects the future to feature a hybrid model combining deterministic and generative UI elements.
Preparing for Consumer AI Agents
Regarding consumer AI agents that execute tasks on behalf of users, Chesky emphasizes that robust infrastructure is required on the platform side. Airbnb aims to become more agent-friendly, viewing external agents as potential lead generators.
Chesky questions whether apps will disappear entirely or simply become data layers under a universal interface. He argues that distinct app designs serve specific functional utilities, meaning apps need to become more agentic while retaining rich user interface controls.
Maintaining deep platform utility requires robust software developer kits (SDKs) or seamless handoffs when external agents interact with specialized features like messaging, map tools, identity verification, and comparative browsing.
Although early industry concerns suggested AI tools might disintermediate companies from customer relationships, practical deployments have shown that conversational interfaces function effectively as lead generators.
Agent Interoperability and Operating Systems
Looking ahead, Chesky envisions agent-to-agent interactions where different software ecosystems communicate directly. While traditional apps required explicit business deals to interoperate, conversational agents can act as interoperable interfaces across platforms.
Airbnb plans to integrate agents into various service areas—from explore tabs to customer support—eventually unifying them into a macro agent capable of connecting with other systems using standards like the Model Context Protocol (MCP).
Despite these advancements, Chesky argues that the broader tech industry lacks a true AI-native operating system. Current AI tools operate within existing environments like iOS, macOS, and Windows rather than serving as foundational kernel-level architectures.
He points out that attempts by platforms like OpenAI to build internal app stores struggled due to the absence of a comprehensive operating system and developer kit equivalent to mobile app ecosystems.
Having tested third-party agent systems, Chesky suggests that consumer AI has not yet been fully realized. Achieving this milestone requires lower-level architectural support, interoperable agent apps, and rich user interfaces.
Misconceptions in Silicon Valley
Chesky identifies two primary misconceptions in Silicon Valley: the assumption that graphical interfaces are disappearing in favor of pure speech and text, and the belief that software will be entirely generative. He contends that human designers remain vital for prompting structured, effective software.
As long as major mobile platforms controlled by Apple and Google remain dominant, standalone applications are likely to retain a central role in consumer technology.
Transitioning fully from apps to agents will require platform-level shifts driven by major operating system providers; otherwise, apps will continue to contain internal agents rather than functioning as fully interoperable standalone units.
As voice interactions gain traction through dedicated dictation and note-taking apps, Chesky anticipates that users will increasingly speak to computers rather than type. Airbnb plans to deploy voice agents to assist users with search and customer service.
Internally, Airbnb has leveraged AI to accelerate feature deployment. On a personal level, Chesky notes that internal AI tools have streamlined information retrieval and reduced his reliance on traditional status meetings.
By reducing communication layers and retrieving data directly via AI, executive workflows become more efficient, decreasing the necessity for frequent large meetings.
