The problem
Most AI projects are planned as a technology change. For the people doing the work, they are a change in who does what. A team that only hears "the agents go live on Monday" fills the silence with its own story: the agent is there to watch them, or to replace them.
So people work around it. They re-check every output by hand, keep the old spreadsheet running beside the new flow, and stop reporting the agent's mistakes because nobody said who to tell. The platform is live, but the work hasn't moved.
That is the thesis: an AI agent joins a team the way a person does. It needs a role, a manager, a probation and a way to be corrected. People accept what they can see and question.
What the AI platform takes on
Each agent should arrive with what a new hire gets.
A job description. One page in plain English: the work it does, the work it doesn't, and the rules it follows. The team reads it before the agent starts.
A manager. A named person who owns the agent's work, answers questions about it and decides when its scope changes.
A visible record. Every action the agent takes is logged with its reason, so anyone on the team can see what it did and why. Suspicion fades when the work is in plain sight.
A way to say "that's wrong". One simple route to flag a mistake, inside the tools people already use, with an answer that comes back. A flagged error is the most useful thing a team can give an agent.
A probation. The agent starts by preparing work for a person to approve. It takes on more only when the record shows it has earned it. This is earned autonomy, and the team watches it happen.
What stays human
The conversation about roles. People hear from their own manager what changes in their job, before the agent arrives, and in person.
Decisions about people. Hiring, roles and performance stay with people. An agent's record is evidence about the agent. It must not be used to judge a colleague.
The time that comes back. Managers and their teams decide together where the freed hours go, and say so early. Silence about this is where the fear of replacement grows.
The right to question. Anyone on the team can hold a piece of the agent's work for review and say why, without needing permission first.
How value is measured
Agree the measures with the team before go-live, and record the current baseline first.
- How much of the agent's output people still redo by hand, week by week.
- How many mistakes staff flag, and how quickly each one is answered. Early on, a rising count is a good sign: people are using the route.
- Whether the old workaround is still running beside the new flow.
- How long a task takes to move from "a person approves every time" to "the agent handles it".
- What the team says in a short monthly check-in: do they understand what the agent does, and do they trust its record?
- Where the hours the agent gives back actually go.
If redo stays high and flags stay at zero, the team hasn't accepted the agent, whatever the dashboard says.
Where to start
Pick one team and one agent. Write the job description with that team, name the manager, and run the probation in the open. The next team will ask how it went. Let the first team answer.
Build, implement and run
Futuvara is a creative technology agency. We build, implement and run AI platforms for our partners.
If AI agents are about to join one of your teams, start a conversation.
