In 30 seconds

An AI workflow lives or dies on whether the people around it actually use it. The ones who do the work today know where the exceptions hide and where a bad answer would cause real harm. Bring them in early, be honest about what the agent is for, and let them shape it. Buy-in is not a launch announcement; it is who was in the room while you built it.

Start with the people who know the work

The person doing the task all day understands its messy reality far better than any spec you write from the outside. Involve them from the start, not to rubber-stamp a finished thing, but to help design it. They will spot the exception that breaks it and the boundary that must never move. This is also the fastest way to a workflow that survives contact with real cases.

Be honest about the goal

People read between the lines. If the real aim is to cut headcount, saying "it just helps you" will not hold, and trust will not survive being found out. If the aim is to remove the dull, repetitive part so people can do the work only a person can, say that plainly, and mean it. Address the worry about jobs directly rather than hoping it goes unspoken; how to handle worries about AI replacing jobs covers this head-on.

Make the first win small and real

Nothing builds buy-in like a narrow, genuinely useful result the team can see. Pick a task everyone agrees is a chore, get the agent doing it well, and let the team feel the relief. A small, real win beats a big promise every time. This connects to how to introduce an AI workflow to a team.

Keep a person in charge

An agent that a team feels was done to them will be worked around. One that has a named human owner, who fields the exceptions and acts on the feedback, becomes part of how the team works. Decide who owns it after launch before you go live.

What this does not solve

Buy-in does not fix a workflow that is a bad idea, and no amount of involvement will rescue an agent people do not trust because it keeps getting things wrong. Bring people in, and also earn their trust with a workflow that actually works.

Good enough to move on when

  • The people who do the work helped shape the agent, not just heard about it.
  • You have been honest about what it is for, including the hard part.
  • There is a first win the team can see, and an easy way to report a bad answer.