Your first AI should interview your people, not replace them
The obvious first use of AI is to take a task off someone's desk. The better first use is to learn how that person actually works. Most of what makes a company run is not written anywhere. It sits in people's heads, in personal spreadsheets, in the email thread where a decision was really made.
The knowledge nobody wrote down
Ask where the knowledge of a process lives and you will be pointed to a manual or a shared drive. Look closer and you find the workaround the senior clerk uses every month-end, the customer who always gets special terms and why, the supplier who must be called before the order is placed. Any system built without that knowledge will fail on exactly those cases.
Let an assistant do the interviewing
An AI assistant can hold a structured conversation with each person about their work, over several short sessions: what they do every day and every month, which decisions they make again and again, who they depend on and who depends on them, where things get stuck, and how they decide when the rules do not give the answer. The output is a written description of the work that people can read, correct and approve.
This does three things at once. It captures knowledge before the people who hold it retire or leave. It shows where the real work is, which is rarely where the organisation chart says. And it gives later automation something accurate to be built on.
Be honest about what it is for
People who describe their work in detail are helping to build the system that will change their job. Pretending otherwise destroys trust the day they find out. Say clearly what the knowledge will be used for, what will happen to their role, and what support they will get, before the interviews start, not after.
Start with the people whose knowledge would be hardest to replace. Their interviews are the most urgent work in the whole project.