Your AI agent can act. Who defines success?

Your AI agents need agreement on what “revenue” means before you give them permission to act.

Sales counts bookings. Finance counts recognized revenue. Operations tracks cash collected.

Ask an AI agent to identify your most valuable customers, and those definitions lead to different answers.

Connect the agent to business systems, and the consequences extend beyond a report.

Two announcements from the last few days address this challenge.

• On November 3, Snowflake announced Business Ontology, designed to give agents approved business definitions linked to enterprise data. Public preview is planned. Its Cortex AI Gateway includes generally available capabilities for tracking usage, controlling spending, and observing activity.
• On November 4, Kong announced expanded AI management capabilities covering agent identity, credentials, cost management, and observability across AI interactions. Kong Inc.

My takeawayApprove the business rules before expanding the agent’s authority.

Start with one workflow.

Assign an accountable business owner.

Document definitions, permissions, spending limits, and escalation rules.

Test ambiguous cases before production.

Keep a human on the loop to monitor outcomes and intervene.

Require explicit approval for consequential actions.

Your AI strategy needs a clear answer to a basic question: does everyone, including your agents, understand the business the same way?