Standardizing MLOps always feels like a trade-off: you either force a rigid framework that kills innovation, or you let the "wild west" of tools create an unscalable mess. One keeps the system stable, the other keeps the data scientists happy, but finding that middle ground is incredibly rare.
Do you think a "universal" framework is actually achievable, or is every team destined to build their own custom patchwork? How do you guys decide when to enforce a standard and when to let the team experiment?