Beyond the CLI: Agentic AI for async workloads and non-developers

Christopher Sanson and Madison Capps from Airbnb challenge common misconceptions about AI, emphasizing its role in augmenting human productivity rather than replacing humans. Airbnb has achieved a 97% weekly adoption rate of agentic AI among engineers without mandates, showcasing organic and widespread acceptance driven by the value these tools provide. The discussion highlights how internal products like AirChat and community enablement play crucial roles in fostering AI adoption, along with the importance of building scalable, modular, and adaptable AI ecosystems inside organizations.

The speakers address the myth that AI has minimal impact on productivity. Airbnb's data reveals a 65% increase in PR throughput and significant rises in coding time among developers using agentic AI. AI tools are enabling developers to spend more time coding and less time in meetings, at the point where over half of the code is now AI-authored. The transformation in developer workflows includes more frequent use of AI sessions, faster migrations, and a shift towards an AI-first approach to product development, radically altering how work is done.

Contrary to the idea that AI coding tools are only for coders, Airbnb's experience shows wide usage across teams, including finance and operations. Non-developers are leveraging tools like AirChat in terminals, building custom applications, and automating routine tasks. This broader adoption is driven by the decreasing cost and increasing capability of AI tools, leading to innovations in product design, migration processes, and data analysis, transforming traditional workflows and expanding AI's reach.

Madison Capps delves into the infrastructure powering Airbnb’s AI efforts, including tools like AirChat CLI, AirChat SDK, and AirChat Remote, which support parallel and asynchronous workflows. The architecture is designed for modularity and scalability, allowing teams to build on and integrate various solutions easily. The vision includes managing multiple agents asynchronously, remote operation capabilities, and setting the stage for an AI-first approach, where work becomes more flexible, scalable, and resilient. The discussion also hints at upcoming developments and the importance of adaptive, community-driven innovation.