Expedia Group's platform engineering lead shares how the team is adapting their large-scale platform to support AI agents as a distinct user class alongside humans. Key insights include: agents trained on open ecosystems struggle with proprietary abstractions, agents will bypass missing APIs by automating browser UIs, and platform teams must design agent-native interfaces. Concrete initiatives include Tarmac (an agent-centric CLI for CI/CD, Kubernetes, and logs), MCP servers for structured platform access, and Koda (a monorepo for agent-built internal developer tools). The broader thesis is that agents are a new class of user requiring dedicated ergonomics, not just a smarter autocomplete layer on existing human-facing tools.
Table of contents
Expedia Group Technology — EngineeringWhen your platform’s next user isn’t humanGet Rick Fast’s stories in your inbox1.8K Impressions