What looks stable in testing starts to strain under load. Latency spikes, costs creep, and systems behave differently as inputs and usage patterns shift. The challenge isn’t just building the system it’s keeping it stable when conditions aren’t predictable. This keynote breaks down what it takes to run AI systems in production, from inference and orchestration through to infrastructure and optimisation. Drawing on real deployments, it focuses on the decisions that determine whether a system keeps working once it’s live. Key takeaways: → What changes when systems move from controlled testing to real-world conditions → The infrastructure and optimisation decisions that impact reliability, cost, and performance → How teams keep systems stable when behaviour isn’t fully predictable