Hi, congratulations on getting to 200 people, and especially on getting non-coders productive. From our own experience of scaling an internal data and AI platform to a couple thousand users, the things that mattered most once adoption passed the first couple of hundred were not the tool itself:
1. Trust in answers. Non-technical users cannot spot a wrong join or a wrong filter, so the quality of the Genie space matters more than the model. Curate each space around one business domain, write column comments and sample questions in the users' own vocabulary, and keep a set of benchmark questions with known answers that you re-run after every change to the space or the underlying tables.
2. Access by default, not by exception. Because people do not know what is happening in the background, permissions have to be right at the data layer. Use Unity Catalog grants, row filters and column masks so a Genie space can never show someone data they could not query directly, and review who can edit a space separately from who can use it.
3. Cost visibility per team. At 200 users the warehouse bill becomes a real number. Decide early which SQL warehouse backs which space, and tag or separate them so you can attribute spend to departments rather than to one shared line.
4. A feedback loop. Ask users to flag wrong answers, and have one owner per space triage those weekly. Most improvement came from fixing metadata and instructions, not from anything clever.
Curious which use case is the most common among the users who are new to Databricks, and whether you have had any wrong-answer incidents yet.
