Choosing
Deciding about AI on an informed basis
Before choosing a solution, you have to know what the AI can do, where it stops, and what its use commits your company and society to. This page exists for that: to inform you, without pushing you either towards enthusiasm or towards fear.
Discuss my constraintsHow far the AI goes
What it can do, and where it stops
The AI reads, summarises, compares, classifies, writes and retrieves at a scale a team cannot handle by hand. Knowing where it stops is just as important as knowing what it does.
What can be handed over to it
Preparation and processing work: going through a pile of documents, extracting the essentials, comparing records, spotting the inconsistency in a file, preparing an answer or a quote, checking information at scale. It saves time on what is repetitive, voluminous or scattered.
It does it all the better when the work is described precisely, the reference documents are up to date, and a person remains able to judge the result.
Where it stops
It does not "know" in the way you do: it produces plausible text, and can assert something false with confidence. It does not carry the responsibility of a decision, does not replace the experience of a trade, and cannot decide for you what binds a person, a customer or a contract.
The boundary is not fixed once and for all: it depends on the trade, the risk and the quality of the checks put in place. It is a leader's decision, not a property of the technology.
The rule we defend
It is not a chat bubble: it is an AI that does a precise job. And as with any worker, you keep the responsibility for what it produces: the human validation points are defined before go-live, never after an incident.
Its involvement in your professions
Deciding the place of the AI in your work
There is no right level for every company. There is the level you accept, profession by profession, after weighing the gain, the risk and the effect on your teams.
Read the project examplesLooking profession by profession
For each job, the question is the same: what happens if the AI is wrong? Preparing an internal note, screening applications, answering a customer or handling a sensitive file do not carry the same consequences. Low-risk, high-volume tasks come first; decisions that affect a person keep a human owner.
The involvement can range from help with preparation — the AI proposes, a person decides — to supervised execution for reversible and well-tested actions. In between, each degree is discussed.
Counting the effect on your teams
An AI installed in a trade changes the daily work: what is prepared for you, what you have to check, what you no longer have to do. Warning people, involving those concerned and training them to check prevents the tool from being something you merely endure. You can also decide not to automate a trade: that is a decision, not a delay.
The stakes to weigh
Choosing with the commitments in mind
An AI choice is not only a technical choice, nor only a budget one. Here are the questions we ask the leaders who ask them — without catastrophising, without minimising.
Jobs and work
The AI transforms posts before it removes jobs: it absorbs tasks, not people. The useful question is which tasks you hand over to the AI and what your teams will do instead. An owned decision is prepared with the people concerned, and measured over time rather than announced.
Personal data and sovereignty
Setting an AI to work on your files means deciding where your data goes, who can read it and how long it is kept. The purposes, the retention periods and the authorised people are validated with the managers concerned; they are not software settings.
Sovereignty is decided here: keeping control of your data is possible. An AI installed on your premises, on a machine in your company's name, is one of the answers — the most direct one when the sensitivity of the data calls for it.
Dependency and real costs
A very convenient external service creates a dependency: prices that move, terms of use that change, a service that closes or degrades. The question is not to refuse the large providers, but to know what it would cost to leave them, and to choose with that constraint in mind.
The real costs go beyond the purchase invoice: installation, operation, updates, training, human validation and the time spent correcting. We put them on the table before the decision, not after.
Resource use
These tools consume energy and machines. The most reasonable answer is not "everywhere, immediately", but "where it really serves". An AI installed on your premises, sized for your real use rather than for a demonstration, plays a part in that frugality: you spend what the work deserves, no more.
Your context
Ask your questions before choosing the solution
Which tools do you use? Which data is sensitive? Who validates the result? What does this change for your teams? These questions will allow us to discuss a project that matches your resources and your ambitions.