Autonomous Agentic Seat Allocation
A travel technology company replaced manual seat allocation with autonomous agents that navigate airline websites using VLM-based UI detection, operating 4-6 concurrent agents continuously.
The challenge
Free seat allocation on airline bookings was performed by human agents working through airline websites manually. The process was slow, error-prone, and could not scale. Direct API access was not available; the only path was through the airline's web UI. Website changes frequently broke existing automation scripts.
What we built
We built an agentic, UI-driven automation system: a Machine SDK managing browser instances and infrastructure, an Agent SDK with standardized behaviors for navigation, form filling, and error recovery, and a Booking Service receiving tasks, dispatching to an agent pool, and tracking status. VLM-based UI detection enabled dynamic adaptation to website changes without script maintenance.
Outcomes
Seat allocation via UI automation without human intervention
Multiple autonomous agents operating continuously
Dynamic adaptation to airline UI changes
Across 5 phases with multi-airline expansion designed
“The agents work through airline websites like a human would, but faster, continuously, and without errors.”