An AI fluent executive assistant can break a task into steps, decide which steps a model should draft, write a prompt that produces usable output first time, and check the result before it reaches you.
Plenty of people can open a chat window. Far fewer can look at a Tuesday afternoon and say which forty minutes of it should never have been done by hand. That judgement is the difference between an assistant who uses AI occasionally and one who is genuinely fluent in it.
Our assistants are trained on the second. They are hired for executive judgement, then trained until using models is as ordinary to them as using a calendar.
Most tasks are a chain: gather, summarise, decide, write, send. A model is excellent at gather, summarise and a first draft of write. It should not own decide, and it should never own send without review. Fluency is seeing that split instantly.
The assistants who get good output are the ones who would also brief a junior well. They give the model the audience, the tone, the constraints, the source material and an example of a good answer. That is why executive experience matters more than technical background.
Sensitive people matters, anything legally binding, anything where being confidently wrong is expensive. A fluent assistant is quick to say that a task is a phone call, not a prompt.
Almost everyone has used an AI assistant to write something. That is familiarity. Fluency is knowing which tasks are suited to a model, how to give it enough context to be useful, and how to check the output quickly enough that the review does not cost more than the drafting saved.
The difference shows up in the second week rather than the first. A familiar user produces slightly faster drafts. A fluent one starts removing whole categories of work.
The single behaviour that separates safe AI use from risky AI use is verification proportionate to consequence. A model is a confident writer and an unreliable witness, so anything factual, numerical or externally visible gets checked against a source.
In practice that means a short rule the assistant applies without being asked: internal drafts go out after a read-through, anything with a number in it is checked against the system of record, and anything client facing is reviewed by a second pair of eyes.
Fluency is quiet. You notice it as briefs that arrive already summarised, meeting notes that separate decisions from discussion, and recurring reports that stop needing a request.
Over a quarter it shows up as scope: the assistant is holding more of your week than they were, without asking for more hours.
Interviews reward people who talk well about AI, which is not the same as people who work well with it. The way around that is to make the interview practical: give a real task with real messiness in it and watch how the candidate approaches it.
A good exercise takes twenty minutes. Provide a short brief, a couple of source documents with an inconsistency between them, and ask for a summary and a recommendation. What you are looking for is whether they notice the inconsistency, whether they verify rather than smooth over it, and whether they can explain which parts of their process they would automate if the task recurred weekly.
Ask them to narrate the second half. The reasoning matters more than the artefact: where did they use a model, what context did they give it, what did they check, and what would they never hand to a model in this task. Fluent candidates answer that easily, because it is how they already work.
Finally, ask what they would document. An assistant who instinctively writes down the working method is one whose value stays in your business rather than leaving with them.
Are your assistants actually trained in AI, or just familiar with it? Trained. Every assistant works through structured training across drafting, summarising, document extraction, scheduling and workflow automation, and is assessed on real executive scenarios before placement.
Do I need to know AI tools myself? No. The point of the role is that the assistant carries that knowledge for you and applies it inside the tools your business already uses.
Will AI output go out without a human seeing it? No. Anything client facing or sensitive is reviewed by the assistant, and anything consequential is approved by you.
More insights | Alvara Partners