Not long ago, the users of an AI application were people. They browsed, paused, and thought between clicks, and every backend built for that traffic inherited its rhythm. Agent traffic has none of it. Agents fan out, retry, and hold state across many steps, bursting without warning and scaling with the value of the work rather than the number of people doing it.
The platforms sitting closest to that shift see it first. Dify sits between builders and the models they ship against, with more than a million applications deployed on the platform. This makes it an early read on how agent workloads behave in production rather than how anyone expected them to. Some of what the team is seeing was not on the roadmap a year ago, and it tends to surface as an infrastructure problem before anyone names it as one.
In this 30-minute fireside chat, Lusha Chen, General Manager of APAC at Dify, and Terry Purcell, TiDB’s Field CTO and Principal Architect, take the move from apps to agents from both ends: what agents are doing that human users never did, and which assumptions in the data layer break first when the user is a machine. This will be a moderated conversation, no slides, with live audience Q&A at the end.
