AI Assistant vs Human Executive Assistant: What I Hand to Each

Thomas Prommer Technology Executive & AI Strategy Consultant Connect on LinkedIn

Published: November 5, 2026

Updated: November 5, 2026

Every executive I talk to now asks some version of the same question: do I still need an EA, or can an AI do it? I have an AI executive assistant with a written charter, and I have handed parts of the job to people too. My honest answer is that the question is framed wrong. The useful split is not AI versus human. It is queue work versus judgment, and it is whether you have written down the rule either one is supposed to follow.

What my AI executive assistant does

I run two AI assistants with separate jobs, which I wrote up in Two AI Assistants With Separate Jobs. Thy is my main executive assistant. Thy owns commitments, travel, follow-ups and one long-running founder thread. Thy reads the channel before answering anything that assumes earlier context, and never answers from memory about a date, a person, a flight or a promise. The charter measures Thy by what is confirmed versus what is still open, always with the source.

The charter is also explicit about the line. Anything that commits my time or money, sends a message in my name, books travel or speaks for me externally gets drafted and then waits for me. Thy does the preparation; I press the button. That one rule is most of what makes an AI executive assistant safe to run every day.

What stayed with a person

Scheduling is the handoff I have rewritten most. In The Delegation Ledger I described how it went: to a person, then to rules, then to a person again. The first time, I handed off the easy half and kept the half that cost me anything, which was saying no. It only stuck once I wrote down the decision rule instead of the preference: protect the first block after training, and decline anything under thirty minutes that could be a message, with a specific alternative.

That rule could be followed by an AI today. What a person still does better is everything around it: reading a relationship before declining, knowing when an important contact gets a yes even though the rule says no, and being in the room or on the phone when it matters. Those are judgment calls with context that lives in people, and that is where an AI's confident wrong answer costs more than doing the work yourself.

Where each one fails

The AI fails on stale context, quietly. A usage table my AI assistant posts showed one of my accounts at 93 percent for twelve hours while the real figure was between 57 and 73 percent, because a cache outlived the login that wrote it. Nobody was careless; the source was frozen. The fix was to show the age of every reading. An AI executive assistant has to tell you how old its information is, or you will make decisions on yesterday's facts.

A human fails differently: on hours and on handoffs. When I wrote about where virtual EAs get hired from, the conclusion was that the choice comes down to which hours the work has to happen in, not cost. A person eight hours away can only take work that survives being answered tomorrow. An AI does not sleep, but it also does not notice when the brief no longer matches what you meant.

How I split the work

• AI takes the queue: inbox triage, follow-ups on quiet threads, first drafts, travel options, keeping a list of what is confirmed and what is open, with sources.
• A person takes the judgment: which commitment matters most this week, exceptions to the rules, anything relationship-sensitive, and presence.
• Neither acts in my name alone: sends, payments, bookings and anything external get my approval, whoever drafted them.
• Both get the same written rules: a handoff with a decision rule compounds; a handoff with only a procedure comes back as an exception.

What changed in 2026

Until this year, running an AI executive assistant meant building one, which is what I did. Between May and October 2026 every large lab shipped an off-the-shelf version: Meta Muse, OpenAI dots, Claude Duet, xAI's Grok Bot and Google's Gemini Spark. They all run on their own cloud computer, check your accounts in the background and ask before they send or pay. Our partner site We The Flywheel compares them in its AI assistants hub.

That lowers the setup cost. It does not change the split above. An agent that can send your email still needs to know which emails you would never send, and that knowledge has to be written down by you first. If you are thinking about this for a whole company rather than one calendar, our partner site CTAIO covers how to roll AI assistants out next to human EAs at enterprise scale.

What I am not claiming

I have not measured whether this saves time. I have success criteria (little stale data, few repeated briefings, actions that actually close) and no baseline yet. Anyone who gives you a percentage for how much an AI executive assistant saves is either measuring something narrow or guessing.

More on this pillar: Compounding CEO.