How Many Agents Do You Have, and Who Owns Them? Agent Inventory Is Enterprise AI's New Baseline

How many AI agents are running in your organization today? Most companies can't answer that clearly. Agents no longer live in one team's single tool; they are born separately inside cloud providers, data platforms and business applications.

The size of the problem

• According to IBM's 'AI in Motion' research, fewer than one in five organizations maintains a complete, current inventory of its AI systems.
• Agents are built on different platforms such as AWS Bedrock, Databricks, Google Vertex, Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce and Snowflake Cortex.
• The pace of creating agents has outrun the capacity to observe and manage them.

The market is starting to respond

On September 24, Dataiku announced Agent Management. The product gathers an organization's agents into a single inventory regardless of the platform they were built on. For each agent it records the owner, what it is supposed to do and what data it can access, measures business and technical performance, and flags the agents that pose the greatest risk. It becomes generally available in October. The same day, a startup called Ando announced a $20M raise for a team chat app where agents join as members with their own identities and inboxes.

Four things an inventory should hold

• Owner: which team and which person is responsible for the agent.
• Purpose: what the agent was built to do and how its success is measured.
• Access: which data, tools and systems it can touch, and with what permissions.
• Behavior: which model it uses, what it costs and how often it fails.

The remot angle

When agents make their model calls through remot, much of the inventory builds itself: which application or agent calls which model how often, how much it spends and which policy applies to it are visible in one place. That way the answer to 'how many agents do we have?' rests on records, not guesses.