From everyone doing their own thing to one shared way of working with AI

At a small software company everyone was already using an AI assistant, but each in their own way. A ninety-minute workshop, built around their own work, laid a common foundation and clear agreements on what can and cannot go in.

The starting point

The team of six was keen to work with AI, and already did. Everyone just did it their own way. Nobody knew exactly what the tool could do, what happened to company data, or what it would cost if the whole team used it. The questions differed per person: finance wanted costs and privacy, marketing wanted an environment without jargon, business development wanted to know how far you could go.

What we did

An introductory session of ninety minutes, in person, in three blocks of about half an hour:

What it is and what it can do. Three kinds of work it is strong at, where it falls short (facts, arithmetic), what licences cost as you scale, and what happens to the data.
How to work together. From loose conversations to a shared environment with shared instructions and ways of working, and a task we built together live.
Limits and next step. What can and cannot go in, the team’s own experiments, and what logically follows.

Every example came from their own work: comparing a client proposal with the standard terms, a monthly report that today runs through an export and a spreadsheet, a reply to a customer with a payment question. Deliberately, nothing was built and no system was connected.

The idea behind it

The gain from AI in a small team rarely comes from six people separately typing better prompts. It comes from a shared foundation: the same expectations, the same agreements on data, and knowledge that does not stay with one person.

So we did not start from a generic demo, but from one simple rule that carries the whole journey: AI proposes, a person decides and executes. Having it act in your own systems is a separate step, set up deliberately, with agreements on paper.

Expected results

After the session the team shares one picture of what the tool can and cannot do, everyone knows how to phrase a task that produces something usable, and there is a concrete list of what may and may not go into the tool. Afterwards the team received the slides, tailored to their situation.

Expected impact

Fewer scattered experiments and less risk with sensitive data. More importantly: a foundation on which a shared workspace and a first pilot project can build, so the team does not start from zero when it wants to go further.