Genie Ontology makes business definitions an ownership question

On November 11, 2026, Databricks announced its next Genie generation. Genie and Genie Ontology are generally available; the new always-on Genie, Search and Ontology Manager still await preview or beta programs. Announcement

That distinction matters for procurement. A product direction can inform a budget without satisfying an acceptance test in today's environment. Open Market Notes sees a more consequential question than whether an assistant keeps working: how will the company maintain the business definitions its AI uses? This is original analysis of public information, not a hands-on product review.

Agreement can conceal the wrong definition

Databricks describes Ontology as combining learned knowledge with human-governed definitions. Product explanation The organizational question is whether a frequently used calculation deserves to become an official rule.

Consider a hypothetical subscription company. Product calls an account active after a key action; finance counts paying customers with valid contracts; customer success checks use during the previous week. These are illustrative examples, not Databricks customer results. All three numbers can be right while answering different questions. Selecting one universal answer might make disagreement less visible without resolving it.

A definition therefore needs a purpose, effective date, owner and exclusions. A reader should know which rule version produced an answer, not merely see a source link. Elevating a team's familiar measure into a company-wide standard should involve business approval. The person administering access is not necessarily the person qualified to decide what a commercial term means.

This need not become a giant dictionary project. Start with a few concepts that trigger consequential actions, such as renewal risk, attributable revenue and qualified opportunities. Establishing who resolves conflicts may be more valuable than connecting another archive.

Certification needs an expiry path

The November release notes put Pages into general availability with certification and deprecation. Release notes For a buyer, approval starts a definition's lifecycle rather than ending it. Reorganizations, packaging changes and reporting decisions can make a formerly correct rule unsuitable.

Downstream outputs need not change together. A central update does not automatically settle what happens to downloaded spreadsheets, presentations or last week's alerts. Teams should distinguish live views from historical records that must retain their original basis. Applying today's definition to an old monthly review can destroy comparability; leaving yesterday's definition unexplained can mislead current decisions.

A useful acceptance requirement is that outputs retain rule versions and calculation times, with original and recalculated values distinguished where necessary. This is a proposed requirement, not a claim that Genie implements the complete mechanism. Test what happens when a definition is withdrawn: does ongoing work stop, request review, or continue under an obsolete rule? The answer can determine whether a customer message or budget change has a sound basis.

A benchmark should inform a test, not replace it

A vendor benchmark developed with Turing reports pass rates of 91% for Genie versus 45% and 58%; comparator agents lacked Genie MCP access. Method and results

This makes preparation of business context a worthwhile variable to investigate. It does not translate directly into deployment reliability or savings. Different tool paths and prepared information affect what the comparison means. Buyers need to understand responses to their own obsolete documents and permission boundaries, rather than paste a public percentage into a business case.

A pilot can combine known-answer questions with deliberately ambiguous variants. Allocate one order by booking, payment and fulfilment date. Leave an obsolete instruction by a former owner linked from several documents. Inspect not only the number returned but whether the system explains its definition, acknowledges uncertainty and stops an unauthorized next action.

Cost evaluation should also include initial curation, dispute resolution, regression testing and incident investigation. Fewer queries can still come with more review work. The investment may remain worthwhile, but those costs belong in the same calculation.

A shared layer can also concentrate mistakes

Databricks plugins allow outside assistants to use Genie's cited answers. Integration explanation Should different assistants therefore depend on the same business vocabulary?

A shared foundation can reduce duplicated configuration and help isolate whether disagreements come from data, rules or wording. But a widely reused wrong definition can affect more workflows. Legitimately different interpretations should remain possible, with their intended scope explicit. Common governance does not require every department to use an identical metric.

Companies with stable data platforms can pilot a frequent, verifiable decision process. Smaller teams that have not agreed on core measures may gain more initially from assigning ownership and writing down rules than from buying additional agent capacity. Automation can help maintain that work; it cannot make management's commercial trade-offs disappear.

Genie's direction merits attention. Connected, certified and safe to act on are nevertheless different states. Useful business AI must explain why an answer applies now, and who will revise its basis when conditions change.