The question we get asked most often, usually about ten minutes into a first call, is some version of: fine, but what an AI executive assistant does all day — concretely. Not the pitch. The Tuesday. What actually happens between 8am and 5pm on a desk that has one, versus a desk that doesn't?
It's a fair question, and most answers to it are useless. The marketing version talks about “productivity.” The skeptical version says it's a chatbot in a trench coat. Neither tells you what lands in someone's lap at 8:15 in the morning.
So here is the honest version, desk by desk. An AI executive assistant— in our case a Claude workspace configured for one person's recurring work — doesn't do a job. It does the first eighty percent of a lot of small jobs, and hands each one back for a decision. What that looks like depends entirely on whose desk it's sitting on.
The office manager's Tuesday
The office manager opens their laptop to 60 unread emails. This is the part of the day that reliably eats an hour and produces nothing anyone would call work.
With an assistant, that hour starts differently. The inbox has already been triaged: sorted into what needs a decision, what needs a reply, what needs filing, and what needs nothing. The obvious replies — confirming a delivery window, acknowledging a resume, telling the copier vendor which Thursday works — are already drafted and waiting in a review queue, written the way this office actually writes.
Then the calendar. Three people need to be in a room, two of them travel, one protects Friday mornings. The assistant holds those constraints, proposes the slot, drafts the invite with an agenda pulled from the last thread on the topic. And the vendor follow-ups — the ones that die because nobody owns them — get written, sent on a schedule, and tracked, so “did we ever hear back from the insurance broker?” has an answer instead of a shrug.
What the human still does: reads the triage, catches the two emails the assistant mis-sorted, rewrites the one reply that needed a human touch, and presses send. Every send is theirs. The assistant does not have the button.
What it gives back: most office managers we set up describe it the same way — the morning hour doesn't vanish, it compresses. Roughly twenty minutes of reviewing instead of an hour of writing, though it varies with how much of the inbox is genuinely new.
The project manager's Tuesday
A project manager's week has a tax built into it: the status report. Everyone already knows what happened on the project. Nobody wants to spend Thursday afternoon converting it into a document a client will skim for ninety seconds.
The assistant is very good at exactly this. It has the messy raw material — the project notes, the threads, the punch list, last week's report — and turns it into a draft status report in the format this client already expects. Same headings. Same tone. Blanks where the PM needs to make a call, flagged rather than guessed.
Meeting notes work the same way. A 45-minute call produces a transcript or a page of scrappy shorthand; the assistant returns a summary with decisions separated from discussion, and every action item attached to a name and a date. Not “we should look into the permit issue” but “permit issue — Dana — by Friday.” The difference between those two sentences is whether the thing gets done.
Client update emails follow. The PM knows what to say. What they don't want is to write the sixth version of “quick update on where we are” this week, in a tone that is warm but not chummy, confident but not overpromising. The assistant produces the first draft; the PM spends two minutes making it true.
What the human still does: everything that requires judgment about the project. Whether the schedule slip is worth mentioning yet. Whether the client is in a mood to hear about a change order. What to leave out. The assistant can draft a sentence; it cannot read the room, and it should not pretend to.
What it gives back: the report that used to eat a Thursday afternoon becomes an edit. It's the most consistent win we see on this desk — not because status reports are important, but because they are relentless.
The bookkeeper's Tuesday
Finance is the desk people assume AI can't touch, because the numbers have to be right. They're half correct. The assistant doesn't do the math and doesn't touch the ledger. It does the writing and the chasing around the math, which is where the hours actually go.
Invoice chasing is the obvious one. Somebody is 40 days late. Somebody else is 70 days late and is also a client you like. Those are two different emails, and writing them well takes more emotional energy than it should. The assistant drafts both — escalating in tone the way your firm actually escalates, referencing the right invoice numbers and dates, and queued on the cadence you set rather than whenever someone remembers.
Then the categorization questions. Not the categorizing itself — that's your accounting system's job — but the pile of “what is this?” charges that accumulate. The assistant drafts the question to whoever spent the money, context attached, so the bookkeeper isn't hand-writing eleven separate “hey, what was this $340 for?” messages.
And at month end, the summary. The numbers come out of the books; the narrative around them — what moved, what's unusual, what the owner should look at — gets drafted in plain English rather than left in a spreadsheet the owner won't open.
What the human still does: owns every number. The assistant drafts commentary on figures it has been given; it does not source them, reconcile them, or vouch for them. If a number is wrong on the way in, it will be wrong on the way out, described fluently. This is the one desk where the review step is not optional and we say so loudly.
The owner's Tuesday
The owner's desk is different, because the owner doesn't have a queue. They have a graveyard: the things that are genuinely important and genuinely never urgent, and therefore never happen.
The proposal you've been meaning to send since last Wednesday. The hiring post for the role you've needed to fill since spring. The follow-up to the person you met at the chamber event who said to reach out. The partner agreement you keep opening and closing. None of these are hard. All of them require a blank page, and a blank page at 6pm is undefeated.
What the assistant does here is remove the blank page. It has your past proposals, so a new one starts from the closest win you've already had, restructured around this client's brief. It has your job descriptions and your actual voice, so the hiring post is a draft to cut down rather than a thing to start. It has the thread, so the follow-up you keep meaning to send is already written and waiting for you to decide whether it's the right week to send it.
What the human still does: decides. Price, scope, promise, whether to take the job at all. The assistant is very good at arranging what you already know. It is not qualified to have an opinion about your business, and any tool that acts like it is should make you nervous.
What it gives back: honestly, less about hours here than about things that now happen at all. That's harder to put on a spreadsheet, and it's what owners tell us about first.
The pattern underneath all four
Read those four desks back and the same sentence sits under each one: the assistant drafts, the human decides.
That is not a limitation we're apologizing for. It is the design. An AI executive assistant is deliberately not autonomous — it does not send, spend, commit, or promise. Every output lands in a review queue with a person's name on it. The human keeps the send button, because the send button is where the liability lives.
There is a real category of work that should run without a human in the loop — a form that routes itself, an intake that files itself, a report that generates on the first of the month. That's a different thing, and we build it under implementation. It has different guardrails and a different failure mode. Conflating the two is how people end up with an AI that emailed a client something nobody approved.
The assistant sits one step back from that line, and earns its keep there: the drafting, the prepping, the chasing, the summarizing, the remembering — the work that surrounds the decisions, which turns out to be most of the work.
Why some assistants stick and others get abandoned
We've watched enough of these get set up to see the split clearly, and it has almost nothing to do with the model.
The assistants that get abandoned are generic. Somebody bought licenses, each person was handed an empty chat window and told it's very powerful. It is. It's also a blank page, and we've established what happens to those. Two weeks later it's a browser tab nobody opens. We've written about this failure pattern in why most small businesses shouldn't start with ChatGPT training — the tool wasn't the problem; the absence of any connection to real work was.
The assistants that stick know the person's actual recurring work before day one. They know that this office manager runs a Monday vendor sweep. They know this PM's biggest client wants the status report in three headings and hates the word “leverage.” They know the owner's proposals always open with the client's problem in the client's words. That specificity is not a nice-to-have — it is the entire difference between a tool someone might open and an assistant they'd be annoyed to lose.
Which means the interesting question isn't really what an AI executive assistant does all day in the abstract. It's what your office manager does all day — the parts that repeat, the parts that are all format and no judgment, the parts that quietly eat Tuesday. Those are the parts worth handing off. For a fast read on where they live in your own business, the free Opportunity Finder is a reasonable place to start looking.
The desks don't change. The work that surrounds them does.
Published by McLean AI Solutions