Snowflake Will Feed AI Agents Only Approved Definitions. Who Approves Yours?
An ontology is only as trusted as the owner behind each term.
Most organizations attribute inconsistent AI agent answers to the model. Upgrade it, tune the prompt, add a better retrieval layer, and the answers should start to line up.
The larger problem is that nobody has agreed what the words mean. On November 3, Snowflake announced Business Ontology, one governed place to record what business terms officially mean, which serves "only approved definitions to agents." In the MIT Technology Review Insights survey cited in the same announcement and sponsored by Snowflake, 55% of executives named lack of business context as the primary cause of low-quality agent insights, ahead of fragmented data at 45%.
That puts a new kind of work on your data team's desk. Before an agent can be served an approved definition of net revenue, someone has to write it, link it to the right tables and sign it off.
And it gets harder as you scale. A handful of agents might lean on a few dozen core terms. Hundreds of agents across finance, sales and operations will lean on hundreds, and every one of them needs an owner who answers when it changes.
As agents become the main users of enterprise data, every agent needs to find the right definitions, the right data and the right other agents, without guessing. That is what we are building at OpenMap: The Agent Network, the connective tissue for the world's AI agents, starting with every person and every local business. We are raising a £5M seed. Meet us at openmap.ai/investors.
Why Does Every Agent Rollout Stall at "What Do We Mean by Customer?"
Picture a sales agent and a finance agent asked the same question: how many active customers did we have last quarter?
The sales agent counts every account with an open opportunity. The finance agent counts every account billed in the last 90 days. Each one is right according to the table it found.
The board sees two numbers.
This is not a model problem. It is an ownership problem.
The Bottleneck Isn't the Ontology. It's the Approval.
Snowflake's design is telling. Definitions imported from other tools arrive as drafts, and owners approve them before agents use them. As dentsu's Joe Tobey put it in the announcement, Business Ontology "gives every definition an accountable owner".
That is the right model. It also exposes a gap most organizations have never closed: for many core terms, there is no obvious owner to approve anything.
Before a single definition can be approved, someone usually has to:
Find every table, view and dashboard that already calculates the term
Work out which version downstream reports and models actually depend on
Trace where the underlying fields come from and how they are transformed
Identify who has the authority to say which version is official
Decide what happens to the reports built on the versions that lose
None of these tasks are particularly difficult. They are slow, they cross team lines, and they usually land on the same few engineers who already know where everything lives.
So the definitions stay in draft, and the agents keep guessing.
How to Get Your Core Definitions Agent-Ready in 30 Days?
You do not need to redesign your data platform, or wait for any vendor's ontology to reach general availability, to start.
Within the first month, focus on the small set of terms your agents and executives ask about most, such as revenue, customer, active account and churn.
Start by mapping where each term is calculated today and which reports depend on each version. Then name one owner per term, with the authority to approve it and the duty to answer when it changes.
Next, expose this context through the catalog or semantic layer your agents already read, whichever platform that is, so the approved definition is the first thing they find.
DataManagement.AI accelerates this process by automatically discovering metadata, mapping lineage across the modern data stack, identifying downstream dependencies, and creating a centralized knowledge layer that stays continuously updated as your environment evolves. Ownership and policy travel with every entity, so when an owner approves a definition, they can see which tables and reports it will touch.
Instead of spending months documenting assets manually before anyone can sign off, your owners start with a map of what already exists. Approval becomes a decision, not an investigation.
Inconsistent Agent Answers Are Actually a Data Ownership Problem
Ontologies, semantic views and catalogs will keep improving, on Snowflake and everywhere else.
But a governed place to store a definition only helps if someone is accountable for what goes into it. Snowflake's Christian Kleinerman framed the stakes simply: "The enterprises that win won't be the ones with the most agents."
Organizations that scale agents successfully do not just publish more definitions. They make sure every definition has an owner who knows what depends on it.
Warm regards,
Shen and Team