Running your own AI stack became an urgent question after the temporary deprecation of Anthropic’s Mythos, and it remains a poor fit for some organizations. This track covers the whole decision, not just the migration: which workloads are worth moving and which are not; what open-weight licenses actually permit, and where indemnification sits; how to prove capability was not lost, by replaying your own logged traffic against candidate models rather than trusting public benchmarks; and how to choose between self-hosting and a token factory, and where the cost curve genuinely crosses.

We will also cover risks, such as; benchmark contamination, dependence on labs subject to other jurisdictions, serving economics propped up by cheap capital, and the alignment and censorship decisions already baked into weights. Sessions span C-level, legal, platform, and AI/ML, because no single function owns the whole path. It’s built for product leads, AI/ML leads, platform engineering and MLOps teams, ideally with at least one LLM workload in production or in pilot. Attendees leave with a transition plan for their own setup.