In short
- The AI lead is a person from the business, not an IT person. They approve what the agent prepares, correct it, report what is stuck and decide with us when the agent gains autonomy.
- Their workload is highest at the start, then should drop as the agent moves up a level. We do not give a figure yet: we will measure it in our first pilot companies.
- A dedicated training course, in preparation, will prepare the AI lead. Getting the team started will be included in every set-up.
In an SME, an AI tool must be run by someone who knows the work it touches. Not necessarily the person who looks after the computers. At Kendu Intelligence, we call this person the AI lead. Here is their job description.
Why a lead rather than an “IT” project
An AI agent is a programme that prepares tasks for you: sorting emails, preparing a reminder, proposing a quote. With us, the agent prepares and your team approves, at least at the start. Its rulebook lists the cases where it must stop. In those cases, it hands over to you instead of guessing.
The real question is therefore not “does the software work?” but “is this reminder right for this client?”. Only a person from the business can answer.
Lack of skills is, besides, the first obstacle cited. Eurostat, the European Union's statistical office, measured it in 2025. Among EU businesses that had already considered using AI, 70.89% cited lack of suitable expertise as a reason for not using it. It was the most cited reason. A trained AI lead is a way to remove this obstacle, in the department concerned.
There is also the European AI regulation (Regulation (EU) 2024/1689, often called the “AI Act”). Its Article 4 is addressed in particular to businesses that use an AI system in their professional activity (the text calls them “deployers”). In its version amended in July 2026, it asks them to take measures “to promote the development of AI literacy” among their staff and the other people who use these systems on their behalf. It states that this obligation does not require them “to guarantee a specific level of AI literacy by an individual”. Appointing and training an AI lead is not a compliance certificate. It is a concrete measure that you can show.
Their tasks: approve, correct, report, decide the levels
Each agent works at an autonomy level, set action by action:
- Level 0, shadow: the agent proposes, your team does the work itself, and we compare.
- Level 1: the agent prepares, a human approves each action before it goes out.
- Level 2: the agent acts alone on the cases written in its rulebook. The human approves the others and checks a sample.
- Level 3: the agent acts alone. The human reads the report and receives the alerts.
The AI lead has four roles.
- Approve. They accept, correct or reject what the agent prepares, in an approval screen in preparation. Each decision will be recorded there in a log: which proposal, which decision, who, when.
- Correct. If the same correction comes up often, it is a badly written rule: the AI lead reports it to us and we update the agent's rulebook.
- Report. A case the agent does not know, software that has changed, a colleague who bypasses the tool: the AI lead reports it to us. A client area, in preparation, will be used for this. Agents will also open a request there themselves when they detect a bug or an error that keeps repeating: the AI lead will see everything in one place.
- Decide the levels. An agent only moves up a level on the strength of the measurements (approval rate, that is, the share of proposals accepted, and number of errors) and with the client's written agreement. The AI lead makes this decision with us. Conversely, an agent will automatically step down one level after a serious alert. And some actions never go beyond level 1: a payment, sending a third party a document that commits you, deleting data.
The right profile: someone from the business, not necessarily from IT
The right AI lead:
- knows the process entrusted to the agent, its rules and its exceptions;
- reads before approving;
- dares to say “this is not right”, to us as to colleagues;
- is listened to in their team;
- has time reserved in their diary, not “when they can”.
They do not need to know how to programme or install a server. That is our job.
Take a typical scenario. An SME with 30 employees entrusts its invoice reminders to an agent. The right AI lead is probably the administration and finance manager. She knows which client always pays late without risk, and which one is in dispute. The IT provider, if there is one, opens access to the software. They do not have to judge each reminder.
Two practical tips:
- Plan a deputy. At level 1, nothing goes out without approval. While the AI lead is on leave, proposals wait.
- One AI lead per business area. If a second agent touches another department (support, for example), take an AI lead from that department.
The time to allow, and how it drops as levels go up
We do not yet have measurements from clients, so no figure. Here is our hypothesis.
| Moment | What the AI lead does | Expected workload |
|---|---|---|
| Shadow trial period, at level 0 (a few weeks, length defined with you) | The team does the work and compares with the agent's proposals | The highest |
| Level 1 | Reviews each action before it goes out | Follows the volume handled |
| Level 2 | Handles cases outside the rules and a sample | Lower |
| Level 3 | Reads the report and responds to alerts | The lowest |
| At any level | Regular review with us, at a frequency defined with you | Occasional |
To check this, the log will record each approval, with its author and time. We plan to derive an estimate of the time spent from it, alongside the other indicators.
If this time does not drop, it is a signal: rules too vague, agent badly tuned, or an action that must stay at level 1. We will discuss it at the review.
Who does what between the AI lead and Kendu Intelligence
| Subject | The AI lead | Kendu Intelligence |
|---|---|---|
| The agent's rulebook | Describes the rules and exceptions, approves the rulebook | Writes the rulebook from our standard agents |
| Build | Provides real examples | Builds the agent and, where their official interfaces (APIs) allow, its connections to your software; tests on anonymised data |
| Day to day | Approves, corrects or rejects | Will monitor, receive the alerts, correct |
| Problems | Reports what is stuck | Handles requests, and eventually those the agent opens itself |
| Levels | Decides with us; your company gives written agreement | Proposes, backed by figures |
| Updates | Takes part in a short trial before going live | Prepares and tests the new version, then gives it a version number |
| Team | Relays, answers simple questions | Trains |
The business owner appoints the AI lead and reserves time for them. Sometimes, it is the same person.
Training the AI lead: content and length
Three training courses are in preparation:
- Getting started, included in every set-up: the team concerned will learn to read a proposal, approve, correct, report.
- AI lead, on quote: the training described below.
- AI in everyday work, on quote: for teams who want to use AI in their daily work, beyond agents.
What the AI lead will learn there:
- Reading their agent's rulebook: what it reads, what it proposes, when it stops.
- Approving with method: what to check, when to reject.
- Recognising a stop case and reacting: a serious complaint, or a person who wants to exercise their rights over their personal data.
- Writing a useful report: what happened, on which file.
- Reading the indicators: approval rate, errors, response times.
- Preparing a level decision.
- Knowing what data the agent may read, and which it must never receive.
- Explaining the agent to colleagues.
Length: defined with you, then adjusted after the first sessions.
A note on funding. In France, since 1 January 2022, a training organisation must be Qualiopi-certified to receive public or pooled funds. This applies to funds from the OPCOs (the skills operators, which fund employee training), the State, the Regions or the Caisse des dépôts. Kendu Intelligence is not yet certified: these funds are therefore not possible for now. Your company can, however, pay for the training from its own budget. Qualiopi is a French certification. For Belgium and Spain, we have not yet checked the funding rules: we will do so with you.
Going further
- Our training, in preparation: getting started, AI lead, AI in everyday work.
- Our method: from the audit to going live, level by level.
- Supervision, in preparation: alerts, report and regular reviews, defined with you.
- Request an audit: we look at your processes and tell you which agent, and which AI lead, would suit you. The audit costs €490 excl. VAT.