The AI Engineer: Designing for Something You Cannot Fully Predict

Deterministic systems fail in ways you can enumerate. Agents do not. Architects are being asked to design for a workload that decides for itself which tools to call and in what order, using patterns built for systems that did what they were told. Most of the standard practice still applies. The parts that do not tend to fail late and expensively.

This session covers what changes when the thing calling your APIs makes its own decisions:

• Where to put boundaries when the workload is non-deterministic
• Why framework-agnostic design matters with the pace of innovation
• How to resource so your engineers can focus on agent logic, not infrastructure

We work through the architectural decisions that are expensive to reverse and where engineers are bottlenecked on agent deployments. Whether you are designing your first agent system or retrofitting governance onto ones already running, this session is about what holds up under load.