Stop the MLOps tools! All you need is SQL
A raft of startups are being created that are building various parts of the new "MLOps" toolkit. Do we actually need them? Perhaps instead of feature stores, experiment tracking and model monitoring applications, all we need is simple SQL? Ethan Rosenthal has written a piece on how a better solution for most people is just to write simple code on top of SQL. No need for dedicated tools.
I wonder how big the market is for MLOps tooling? Some smaller or non-tech companies have a few models and can build and manage their own set-ups. Some companies have ML as their key capability and want to be in control of everything. Big tech firms have thousands of models and are too big and need their own custom infrastructure. That leaves medium sized companies for whom ML is not central to their mission. My guess is they might want to go for open sources frameworks over custom commercial tools?