When is a surprise a good thing?
On your birthday. At a hotel front desk, when they hand you the keys to a better room than the one you paid for. Those are great. Somebody thought about you in advance, and the surprise is the proof.
A surprise in the middle of a transaction is the opposite. It means nobody was thinking in advance. It means nobody was watching the file.
What does that surprise usually look like?
A signature that was never collected, found the day before closing. A disclosure sitting in someone's inbox. A deadline that moved two weeks ago while everyone was heads-down on the part of the deal that felt urgent at the time.
Every one of those is ordinary work that went unwatched. It shows up late because nothing was tracking it. That's why great transactions aren't improvised. The good ones look calm from the outside because somebody did the anticipating early, when it was cheap.
So what should an assistant actually do?
Three things. What's missing. What's next. And what could become a problem before it does.
That's the whole job. Not writing your listing copy, not sounding clever in a chat window. Just holding the state of the deal in view, all the time, and saying something while there's still room to fix it. The moment a problem has to be handled in a fire drill, the assistant already missed its window.
What does that rule out?
Anything unpredictable.
An assistant you can't predict isn't helping you. If you have to check its work, you're doing the work twice, and you're carrying a new kind of risk on top of the risk you already had. Unpredictable AI isn't an assistant. It's just chaos with a login.
How would you know it's working?
Nothing happens.
That's an uncomfortable thing to sell, because there's no moment to point at. No save, no dramatic catch, no story to tell at the closing table. The deal just moves, and the things that would have blown up got handled so early that nobody remembers they were ever risks.
Boring is the goal. Save the surprises for the birthday.