The unspoken question
When a team is told an AI teammate is joining, everyone privately asks the same thing: is this replacing me? If leadership doesn’t answer it directly, the team answers it for themselves, usually pessimistically, and cooperation quietly disappears.
You need their knowledge to make it work
The system’s quality depends on the decision rules that live in people’s heads. Those only get written down by people who want the project to succeed. A team that feels threatened will withhold them without ever refusing anything.
The knowledge that makes an AI employee good is owned by the people most worried about it.
Be specific about scope, out loud
Vague reassurance reads as evasion. Naming exactly which tasks it takes and which it does not is both more convincing and more useful, and you have already written it if you produced a job description for the role.
Give the team authority over it
The strongest signal is putting the team in charge: they review its output, they decide what it handles, they can switch it off. That converts the relationship from being measured by it to managing it — and it is also just the correct ownership structure.
Start where the work is disliked
Deploy it first against the part of the job nobody enjoys. Enthusiasm is easy to earn when the first outcome is less tedium, and it buys the goodwill needed for the corrections the first month requires.
Say what happens if it works
If capacity is freed, say what it gets spent on. If roles change, say how. Ambiguity here is what people fill with the worst case, and it is the difference between a team that reports problems early and one that lets you find them from customers.