AI agent costs follow usage, not named-user licences
In brief
According to Journal du Net, most large groups let IT fund generative AI and charge the cost back to subsidiaries by number of users. That fits conversational assistants tied to a person. It does not fit agents, which consume tokens, API calls and supervision time for a business function rather than for one employee.
Source : Journal du Net - « Agents IA : pourquoi la refacturation par utilisateur ne tient plus »
What it changes for an SME
A firm of 10 to 250 people rarely charges costs between subsidiaries. The issue still matters: if all AI spend sits in IT, no business owner is accountable for an agent's results. Journal du Net notes that an agent serves a function such as accounts payable or support, not a named user, so usage needs to be measured if the cost is to sit with the team that benefits. According to the Oliver Wyman Forum (The Reality Check, October 2026), as cited by Journal du Net, one hour of AI-assisted work comes with 22 minutes of checking on average; that time is part of the real cost even when no invoice shows it.
What you can do
- Before the first agent goes live, decide whether IT or the business function that uses it will pay.
- Check whether your AI tools bill per seat or per use, and track actual consumption rather than license counts alone.
- Include human review time in the cost, and do not treat an agent as a full-time job you can simply remove.
Let's talk about the time you could save.
In 20 minutes, we find where AI could save you time or money. Free, no commitment.