AI agents can investigate data and recommend decisions, but they need a company’s definitions and rules to make their answers useful. Zoho’s Raju Vegesna argues that organisations should build that context into governed data, shared definitions and clear ownership, rather than expecting an agent to guess what “revenue” means.
Watch Desk analysis
What happened
In a 5 November article for Diginomica, Vegesna traces business intelligence from IT-built reports through self-service tools, augmented analytics and generative AI to today’s agentic systems. His central point is that agents need more than access to company data: they need to understand how the business is organised and what its terms mean.
He describes two building blocks. An ontology maps a company’s core business units and how they relate; a semantic layer defines metrics, synonyms and instructions. Together, these can give agents a usable guide to business meaning. For example, sales may count revenue when a contract is signed, while finance recognises it only after obligations are met. An agent needs to know which definition applies, and what to do when a question does not specify.
Vegesna recommends assigning owners to definitions, recording versions and changes, and logging which agents accessed which data. He also points to the Open Semantic Interchange, launched in 2025 by Snowflake, Salesforce, dbt Labs, RelationalAI and BlackRock, as an example of work towards shared data definitions. Read the article.
Why it matters
An agent can produce a fluent answer and still use the wrong definition. In a business, that is not a philosophical quibble: two teams can use the same word for different measures, and a system that silently picks one may give decision-makers false confidence.
The practical takeaway is to start with the measures agents are most likely to use, name who owns each definition, document the rules and changes, and make access and activity traceable. The article also cites Gartner forecasts about agent adoption and governance failures; those are forecasts, not a guarantee that any particular company will follow the predicted path.
Our read
This is the unglamorous work that makes an agentic demo fit for a real organisation. Before adding another agent, settle what the important business terms mean, who gets to change them and how those changes reach the systems using them. It is not a dazzling launch, but it may save a dazzlingly confident answer from being confidently wrong.
What to watch
- Whether companies assign named owners to their most-used business metrics.
- Whether semantic layers and shared standards become routine parts of enterprise AI systems.
- Whether organisations publish evidence that governed definitions improve agent answers in practice.
Discussion spark: Should business teams or data teams have the final say over the definitions an AI agent uses when their numbers disagree?
Sources and evidence
- Want to hear more about Agentic AI? (5 November 2026, 13:30 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.