AI Reliability: Bridging Frontier Scale and System Stability

Frontier AI breakthroughs grab headlines, but real success lies in whether systems hold when millions query at once, GPUs fail mid-inference, or upstream dependencies degrade. Inspired by Rootly’s Top People Making the World More Reliable, this session unpacks the operational lessons of 2026. Join infrastructure leaders and SREs to explore practical architectures that prevent cascading failures, automate incident response, and turn non-deterministic AI into dependable enterprise software. Key Topics: Cluster & Compute Resilience: Mitigating GPU node failures, memory leaks, and distributed training/inference interruptions. Inference Under Pressure: Designing fallbacks, caching, and circuit breakers for real-time model and agent workflows. Redefining AI SLOs: Measuring uptime, latency tails, and reliability in probabilistic systems. Postmortems in Practice: Blameless breakdowns of 2026’s high-impact AI outages and architectural takeaways. Who Should Attend SREs & DevOps Engineers: Teams managing distributed cloud scale, high availability, and multi-nines uptime. MLOps & Platform Engineers: Engineers building inference pipelines, model serving platforms, and GPU orchestration layers. Incident Commanders & Ops Leads: Leaders guiding Sev-1 response playbooks, runbooks, and on-call practices. Technical Leaders (CTOs, VPs, Directors): Decision-makers balancing rapid AI deployment with platform stability and customer trust. Agenda Check-in & Networking: 5:30 p.m. - 6:30 p.m.