Belgian Data Guy/gent
← All posts2 March 2026

Engaged by data, not overwhelmed by it: what Lakebase and Genie Spaces change

Almost every client I walk in on already has more dashboards than they have questions. Someone asked for a report eighteen months ago, it got built, and now it sits in a folder nobody opens because the actual question this month is slightly different. The business isn't short on data — it's overwhelmed by an ever-growing pile of pre-built answers to questions it no longer has, with no easy way to ask a new one without filing a ticket to the data team.

Genie Spaces attacks that from the access side. It's a conversational layer Databricks puts on top of a governed set of tables, letting someone in the business type a question in plain language — 'which product lines had margin drop in Q3 compared to plan' — and get back a real, governed query result, not a canned chatbot guess. The part that actually matters here isn't the chat interface, it's what sits underneath it: Genie only gives good answers over a small, well-curated, pre-joined set of tables. Point it at your entire raw warehouse and it will confidently misread ambiguous column names the same way a new analyst would. Point it at a handful of clean, purpose-built views and it turns a two-day reporting request into a two-minute question.

Lakebase attacks the same problem from the plumbing side. Historically, if you wanted an operational app or an AI agent to read and write against live business data, you'd stand up a separate Postgres instance and build a pipeline to keep it in sync with the lakehouse — one more copy of the data, one more thing that drifts out of date, one more source of the 'why don't these two numbers match' conversation that erodes trust in data generally. Lakebase folds that operational, transactional layer into the same governed lakehouse environment, so the AI agent or the internal app is reading the same governed data the Genie Space and the BI dashboard are, instead of its own quietly diverging copy.

Neither of these is a shortcut past the modelling work — if anything, Genie Spaces raise the bar on it, because a business user asking an open-ended question in plain English will hit every ambiguity in your semantic layer that a fixed dashboard used to paper over. Where I've seen this land well is when a platform already has its dimensional layer in decent shape, and Genie becomes the front door for the long tail of ad-hoc questions that never justified a dedicated dashboard, while Lakebase quietly removes the extra synced copy of the data that used to sit behind every 'operational' app. That combination is what actually turns a data platform from something the business has to read into something it can just ask.

Working on something similar?

I take on a small number of data platform engagements at a time.

Get in touch