About · Kendu Intelligence

My background: I have been in your shoes

In short: I learned to sell and to count, then I was a business owner. Today, I put that experience at the service of SMEs.

The journey, in four stages, up to Kendu Intelligence.

Selling and counting

I graduated from EDHEC (BBA), SKEMA (international marketing and business development) and Audencia (MSc for finance directors). I kept two outlooks from it. The first: how a company sells and grows. The second: what each decision costs and earns.

Hands-on, in the United States

As an expatriate in the United States, I worked at the same time in a wine shop and in an import and distribution company. On one side the shop, on the other import and distribution.

Business owner, from recruitment to software

I was a business owner. First in international recruitment. Then, for several years, in SaaS software: those online programs you pay for by subscription. I saw two opposite approaches. Recruiting is finding the right person for a specific job. A SaaS is the same software for everyone.

Today, custom work

I trained in current technologies to put this experience at the service of business owners. My aim: to help them improve their performance with personalised tools, rather than generic ones.

That is what Kendu Intelligence offers to SMEs in France, Belgium, Spain and Italy, in three forms. AI agents: programs that prepare repetitive work in your tools (payment reminders, accounting documents, emails, quotes). Custom software. Training.

What I took from it, and what you gain

In short: a tool should start from your daily work, prove itself in a trial and leave you the decisions that matter.

Start from your daily work. In the United States, I saw two different trades up close: the shop, and import and distribution. I learned that a useful tool starts from the real work of the people who use it. So Kendu's agents prepare repetitive work where it happens: in your tools.
Judge on a trial, not on a promise. Recruitment taught me that a recruit is judged at work, not only on a CV. I apply the same rule to agents: each is “hired on a trial basis”. For a few weeks, it suggests, your team decides, and we measure. Then its autonomy rises level by level, that is, step by step, with your written agreement.
Count what it earns. My finance training taught me a simple question: what does it earn, compared with what it costs? For a tool, I prefer to measure than to assume. That is the role of the trial.
Fit the tool to you, not the other way round. SaaS showed me the strength of shared software, and its limit: each company has to adapt to it. Custom software does the opposite. The tool starts from the way you work.
Respect your time and your decisions. I was a business owner. I know that a manager's time is limited, and that they answer for what binds their company. The agents prepare the work. But paying, sending a document that binds you or deleting data always goes through a person.

How I work with AI: I decide, checks verify

In short: the software is written by AI agents. I decide what we build and what we ship, and nothing goes out until the automatic checks are all green.

What to build. Before any work, we write a specification: what the software must do, for whom, and what it will not do. An agent tasked with challenging it reads it first and points out its weaknesses. I read the specification, its objections and the answers given. Then I approve it, or not. Nothing is built before I agree.
Whether to ship. At the end, I receive a one-page summary, in plain language: what changes for the user, where to see or test it, what was checked, what remains to be done. I say yes or no on that summary. Without my go-ahead, nothing goes live.

A plan. It splits the work into steps, says what we are not building and gives the expected size. If it goes beyond what the specification provided, it goes back to the agent that challenges.
The build. The agents write the code one step at a time. Each step is checked before the next: tests, code checks, building the site.
Reviews that try to break things. Everything sensitive (access, payments, personal data, the database, money) is reviewed by two agents. Their role: to try to make the work fail, not to approve it politely.
A final check. The result is compared with the specification. If a gap is legitimate, we correct the specification. Otherwise, we correct the code.

the software's tests;
code analysis;
a search for passwords or keys left in the code;
an audit of external components, those software building blocks written by others that we use;
security checks. For example, when software serves several customers, a check makes sure none can access another's data.

Key facts

2decisions from me: what to build, and whether to ship
4steps chained together by the agents
4target countries: France, Belgium, Spain, Italy

What I decide

The work moves forward in batches, that is, in pieces delivered one after another. On each batch, I make two decisions.

What the agents do

Once the specification is approved, the agents chain the steps together.

What is checked, without me

I do not read the code. My control lies elsewhere: in what we build, and in what is verified. That is why the automatic check is blocking. Until it is entirely green, nothing is delivered. It includes:

Even a small change goes through the tests, a review and my go-ahead before going live.

What this does not guarantee

An AI agent can make mistakes. An automatic check only sees what it has been taught to look for. The process reduces this risk; it does not remove it.

This is one of the reasons custom software is delivered after acceptance testing, that is, a point-by-point check with you. It is also why our agents start by suggesting, without acting.

This process is written down and kept up to date. Consistency between the documents and the code is re-checked at regular intervals.

Let's talk

Does a repetitive task take up your time every week? Does a tool not fit the way you work? Write to me at contact@kendu-intelligence.com. Let's simply talk about what takes up your time today. Then let's look together at what a tool made for you could change.

AI agents designed for SMEs, by Maxime Desbonnet