AI agents: a wide scope raises cost and risk for SMEs
In brief
According to Journal du Net, an SME planning an AI agent should first define the work that stays human. Covering several jobs at once makes concrete testing hard, slows down finding why an answer was wrong, and can undermine staff trust. Three filters help: how often the task occurs, whether the rules are stable, and what an error would cost.
Source : Journal du Net - « Agent IA en PME : commencer par ce qu'il ne faut pas automatiser »
What it changes for an SME
For a firm of 10 to 250 people, the practical change is to start with a single task rather than an assistant that spans several jobs. A rare task, unclear rules, or a mistake that would reach a customer or commit money all argue for human review, or for waiting. When the agent talks to the public, Journal du Net notes that Article 50 of the EU AI regulation, applicable since 2 August 2026, requires people to know they are dealing with an AI unless the context already makes that clear. On cost, a market tool tried for a few weeks on real cases should be compared with a custom build over the same period, including maintenance.
What you can do
- Write down the work you will not give to an agent, then keep only one frequent task with stable rules and an error you can correct internally.
- Before connecting any system, ask for the processor contract required by Article 28 of the GDPR and check where personal data will go.
- If the agent will speak to customers, check that they can tell they are addressing an AI, as required by Article 50 of the EU AI regulation.
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.