I need someone to double check me here because I'm hoping I'm confused. So Databricks has it's nice no-code data classification tool in the UI to let users orchestrate AI functionality. Very cool so far.
However, one small problem - there's no option in the UI to select a model. It just runs the default offering from the model reg. So I think maybe that's just not implemented yet and go the SQL route to use ai_classify() directly. Lo and behold you can't choose the model here either, and the built in model choice is still opaque! What? Databricks doesn't seem to do any analysis on my workflow to automatically route to an appropriate model. Is it really just using the same model regardless of prompt complexity?
If this is the case, what is the actual usecase for this functionality? I have been cheerleading Databricks at work for a while now and have onboarded a ton of users, but how can I endorse using ai_classify() in workflows when the spend has the potential to be so wildly mismatched to the task?
I'm currently advising people who want low code agentic solutions to use n8n, which is a shame because pretty much all the other resources they might want to include are in Databricks. Of course technical users can use ai_query(), but the overlap of people who want to code their own agentic workflows and people who want to use Databricks built-ins is pretty small.