Humans Are the Most Unreliable Node in Your System

Modern software systems are built to handle unreliable networks, partial failure and distributed state. But there’s one non-deterministic component most architectures fail to model properly: the human. Users retry actions mid-flight. They interrupt firmware updates. They unplug devices under load. They optimise for speed over safety. And when software interacts with physical systems, those behaviours become architectural risks. In this session, we'll dissect real-world failures from connected products and explores how unpredictable human behaviour cascades into state corruption, retry storms, inconsistent device-cloud views and recoverability problems. You’ll learn how to: - Model human behaviour as hostile or non-deterministic input - Design state machines that tolerate interruption and repetition - Make critical flows idempotent across device and cloud - Build systems that degrade safely when users don’t behave as expected If your software leaves the browser or the data centre, this talk reframes reliability as a human systems problem, not just a technical one.