Preparing your teams for the arrival of AI agents

The lack of in-house skills is the obstacle most often cited by heads of French SMEs and mid-sized companies when it comes to AI. But the most frequent obstacle on the ground lies elsewhere: a team that was not involved works around the tool, and the best agent in the world brings in nothing. Here is how to avoid that.

Involve the person who does the work, from scoping onwards

They know the exceptions — and a process is won or lost on its exceptions, not on its routine case. Involving them from the start has two effects:

• the agent is better, because it handles the real cases;
• the agent is accepted, because the person helped build it.

We even ask, where possible, for them to attend the Flash Diagnostic interview.

Say what changes, for each person

Concerns come mainly from vagueness. Say clearly:

1. what the agent takes on — and what it does not do;
2. what the person does from now on: approve, handle exceptions, supervise;
3. what the time freed up will be used for. In an SME, it first goes to what had been put off for lack of time.

Make the control rules visible

A team accepts an agent more readily when it knows who decides. Show the three modes:

• the agent acts alone on what is reversible and low-stakes;
• the agent prepares, a person approves before it goes out;
• the agent prepares, a person decides and executes what cannot be undone.

See Can an AI agent make mistakes?

Organise real onboarding

A 60 to 90-minute session, with the people who will use it, on real cases:

• what to ask the agent, and in which cases;
• what not to hand over to it;
• what to do if it gets something wrong, and how to report it;
• where to check its status.

A simple sheet, given to each person, sums it all up. It is included in our Process Agent and Business Agents projects.

Appoint a lead

A person who knows the business, reports badly handled cases and keeps the rules alive: that is the role of the internal lead. They do not need to be technical. See The internal lead.

Measure together

Share the agreed indicator — time spent, processing time — before and thirty days after. A visible gain, measured on their own work, convinces better than a speech.

Training: an obligation, in proportion to use

Since February 2025, the European AI regulation requires the people who use AI to be trained, in proportion to their use. For a specific agent, onboarding contributes to this; for broad use of off-the-shelf tools, a usage charter and some familiarisation are useful. See AI Act: what changes for SMEs and mid-sized companies.

Sources

• Bpifrance Le Lab and Numeum: the lack of in-house skills cited by 60% of French SMEs and mid-sized companies as an obstacle to adopting AI.

Frequently asked questions

Will AI agents cut jobs?

They take on tasks and processes, not whole jobs. In an SME, the time freed up first goes to what had been put off for lack of time: following up customers, preparing quotes better, handling exceptions.

Should the whole company be trained in AI?

Everyone should be trained in proportion to their use — this is in fact an obligation under the European AI regulation. For a specific agent, an onboarding session with its users is usually enough.

What if an employee refuses to use the agent?

Understand why. A refusal often reveals a badly handled case, a forgotten rule or a legitimate concern. It is valuable information for improving the agent.