Most teams meet AI as a tool. You open it, you type, it answers. That framing is comfortable and it is wrong, because a tool does what it is told and this does something else: it reads the situation, it fills in what you left out, and it decides how much to decide.
So the first question is not what the model can do. It is where it should sit in the work. There are three seats, and a team that can name them steers.
Execute, contribute, defer
Execute is the seat for work with a known shape. A brief with a fixed structure, a translation into a format the team has agreed, a first pass at a summary. The model runs, a person checks, and the check is fast because the shape was fixed before the run.
Contribute is the seat for work that is still being thought. The model drafts an angle, proposes a cut, argues the other side. A person is in the loop the whole way, because the value is in the exchange and not in the output.
Defer is the seat for work the team keeps. The judgment call, the relationship, the decision that carries a name. The model can prepare the ground, but it does not sit in the chair.
Mapping the work, not the tool
Navigating means mapping every recurring piece of work onto one of those seats, in the open, with the people who do it. It is unglamorous and it is the whole game. Once the map exists, adoption stops being a question of enthusiasm and becomes a question of routing: this runs, this is argued, this stays ours.
The map also tells you what to encode next. Anything in the execute seat wants a fixed shape written down, so that the next run is as good as the last. That is the second note.
