Will This Replace My Office Manager?
The honest answer is no. Sometimes the honest answer is you should not buy this.
I have an interest in answering this badly. If I told an owner the machine would take the office manager, the close would be cleaner and I would sleep worse, which is a lousy reason to lie. I ran this on my own firm first, at B:Side Capital, and after we built MARCUS, fourteen agents across seven departments, from about 840 of our own process files, the people who already knew the work were still there.
The answer is no.
What the ninety days returned
In the first ninety days of full-fleet operation it returned 1,240 staff hours of preparation, a run rate of about 5,000 hours a year, or roughly 2.4 people of pure prep. That figure is prep, not a headcount we cut, and nobody was fired so a model could do the gathering. The same people reviewed the work they used to assemble, which is the whole point of giving the hours back.
By week six, ninety-three percent of the staff were using it weekly, and the first real question arrived eleven minutes after go-live. Every consequential action still waited on a person. Nothing sent, filed, posted, or paid until someone approved it, which held one hundred percent of the time. Eighty-nine percent went through on the first pass, nine percent came back with comments, and two percent escalated.
Silence never means yes.
If the office manager can't question the draft, the draft is unowned, and the shop has only gotten faster at being wrong. The brief we wrote for ourselves was one line: take the prep off our people's plate, without ever taking them out of control.
The job that stays
The future doesn't belong to the people who can do what AI does; it belongs to the people who can take what the machine produces and be accountable for it. A human has to be able to explain the work, question it, and overrule it. If you run a twelve-person practice, that human is usually the office manager, because the machine can stage the file and she still owns the call.
There's absolutely nothing wrong with wanting the hours back, and I wanted them too, even when the attractive version of the sale was a seat that replaces a salary. That version fails the first time a draft is wrong, because the person who used to catch the mistake is gone and the model doesn't know it made one.
Sometimes the honest answer is you should not buy this. The right move is often to wait a quarter, watch one workflow, and keep the person who already knows the work. A quarter spent watching one workflow is cheaper than a year spent defending a tool nobody will overrule. Most shops shouldn't buy a fourteen-agent back office, so start with one job the office manager already hates, and a rule that she can kill it.
I don't know yet how every twelve-person practice should copy what we did at B:Side, or whether they should. That part is still open.
If you're carrying a question this page didn't answer, that's what the free assessment is for. Thirty minutes, a senior advisor, no deck.
Where this shows upWhat we actually build →
