My line on this has been the same for years: automate before you are automated.
It is not a threat. It is an observation that the parts of this job that a machine can do are going to be done by a machine sooner or later, and the assistant who has already handed those parts over is in a very different position from the one who is still defending them.
What has changed recently is that the tools are finally good enough to be worth the trouble. What has not changed is which parts of the job actually matter, and almost none of them are the parts AI is good at.
Start with what does not transfer
Before the tool list, the part that gets skipped.
An assistant's real value is judgment, relationships and context. Knowing that this request from this person on this day means something different than it looks like. Knowing who to call. Knowing that your executive says yes to things on Friday afternoons that they regret on Monday.
None of that is in a model. It is in you, and it is built by being present for years. Fiona Young on AI and the future of the EA role is the clearest conversation I have had about where the line falls, and Dawn Stallwood on automation and leadership gets at the same thing from thirty years of watching leaders.
So the goal is not to do your job with AI. It is to hand over the part of your job that was never the point, and spend the returned hours on the part that was.
Where it genuinely helps
From the conversations, and from my own week, the wins cluster in four places.
Drafting anything that has a shape. Meeting agendas, briefing docs, a first pass at an email you will rewrite anyway, notes turned into a summary. The value is not that the output is good. It is that a bad first draft is far easier to fix than a blank page.
Summarizing volume. Long email threads, recorded calls, a fifty-page board pack you need the three relevant pages from. This is the single highest-return use for most assistants, and it is low-risk because you can check it against the source.
Research you would otherwise not do. Background on a person before a meeting, an unfamiliar industry term, three options for a venue. Verify anything you will act on, and treat the output as a starting point rather than an answer.
Removing the repetitive middle. Not the thinking, the plumbing. Moving information between systems, filing, formatting, chasing. Cortney Hickey, executive business partner to the CEO of Zapier, talks about scaling a role this way, and Beth Portesi builds her own small tools for this after twelve years as an EA, which a year ago would have required a developer.
Where it does not, and where it is dangerous
Anything where being wrong is expensive. Numbers going to a board. Names and titles in an external email. Dates. A model that is right ninety percent of the time is useless for a task where the tenth time costs you your reputation.
Anything that is really a relationship. The message that needs to sound like your executive, to someone who knows them. People notice, and the cost of being caught sending generated warmth is much higher than the time it saved.
Confidential information, until you know the rules. This is the one that gets assistants in real trouble. Board material, personnel matters, pre-announcement financials, anything covered by an NDA. Find out what your company's policy actually is before you paste anything into anything, and if there is no policy, assume the answer is no and go and ask. You have access to more sensitive material than almost anyone at your level, which is exactly why this matters more for you than for a colleague experimenting with the same tool.
How to actually start
The assistants who have made this work did not begin with a tool. They began with a task.
- Pick one recurring task you do at least weekly, that takes more than fifteen minutes, and where being wrong is cheap. Meeting notes into a summary is the usual first one.
- Do it both ways for a fortnight. Your way and the tool's way. You are finding out whether it is actually faster once you include the checking, which is the part everyone leaves out of the estimate.
- Keep what survives. Most experiments do not. That is a successful experiment, not a failure.
- Then add a second task. Not a second tool.
Mike Todasco on leveraging AI as an assistant covers the landscape from someone who ran innovation at PayPal, and Naomi White, a senior EA at a company that builds these models, talks through what she actually uses them for.
If you want to be walked through it rather than reading about it, Beth Portesi and I ran a workshop for assistants on putting this into a real workflow.
Getting better output
The single biggest difference between assistants who find these tools useful and those who do not is how much context they give.
A one-line request gets a generic answer. What works is telling it who you are, who the output is for, what the constraints are, and what good looks like – roughly what you would tell a capable new team member on their first day.
Two habits worth building:
Keep your good prompts. When something produces a genuinely useful result, save the wording. Most of the value of a year with these tools is the dozen prompts you end up reusing weekly.
Give it your examples. Paste in three agendas you wrote, then ask for a fourth in the same style. Output that sounds like you is almost always a context problem, not a tool problem.
On the job question
Some version of "will AI replace assistants" comes up every time I speak anywhere, so plainly: I do not think so, and I have not for years – episode 149 is the chapter where I laid out why, and I wrote the same argument down in five ways to future-proof your career back in 2019, before any of this was fashionable.
What I do think is that the role keeps moving up. The scheduling and the inbox-sorting get automated, the judgment and the partnership do not, and assistants who lean into the second half keep getting more valuable. I have written a more current version of this in how to future-proof as an executive assistant in the age of AI, and Leah Warwick and Maggie Olson on the state of the profession is a good read on where the whole field is going.
The assistants I worry about are not the ones whose tasks get automated. They are the ones whose entire job is tasks.
A note on specific tools
I have deliberately not listed products, versions or prices here. That list is wrong within months, and a guide that ages badly is worse than one that stays general.
What I will say is that the tool matters much less than the habit. Assistants getting real value out of this are not the ones who picked correctly. They are the ones who picked something, used it every day for a month, and found the three things it was genuinely good at.