Charging AI agents by user is outdated

In many large corporations, generative AI tools are managed by the IT department and billed to subsidiaries based on the number of users. This model works well for conversational assistants but is less effective for AI agents, which perform specific tasks rather than serving individual users.

AI agents work differently from assistants

An AI agent is not assigned to a single person but operates for a specific function, such as supplier accounting or recruitment. It often interacts with other agents and consumes resources like tokens and API calls rather than a fixed subscription. This makes traditional user-based billing models less suitable.

The AI Gateway tracks agent usage

To better manage costs, companies are implementing an AI Gateway, a central system that tracks all calls to models and agents. This allows organizations to trace which agent consumed what resources and link the cost back to the specific function or department that initiated the request.

Basing costs on functions improves accountability

By attributing the cost of an AI agent to the function that uses it, organizations can hold business managers accountable for its performance. This contrasts with a general IT budget, where costs are shared and no one is directly responsible for the agent’s efficiency or impact.

Charging by usage has limitations

While billing by usage is more accurate, it is not a perfect solution. AI agents do not always replace full-time positions, and their real cost includes human supervision time, which is often overlooked. These factors must be considered for a fair and transparent billing model.

Organizations must decide who pays for agents

The question of who should bear the cost of AI agents—IT, the using function, or no one—has not been widely addressed. Companies should consider this decision early in the deployment process to ensure accountability and proper governance of AI usage.