The executive, the inbox, and AI

What is your relationship to your inbox? If you're like most of us, the answer is probably "it's complicated".

This simple message channel is about more than sending and receiving information. It dictates our workload, acts as a gatekeeping mechanism, and signals status and responsiveness. For many it's experienced as a dangerous time sink, resented but inescapable.

I'm not here to tell you that AI will magically change your relationship with your inbox, because, well, it's complicated. In fact there's an argument to be made that a better email written by AI is actually worse at communicating the intended felt message than one with authentic typos throughout.

But there are ways you can use AI to improve your relationship with your inbox. These aren't about saving time, although you will. They're about helping you decide what gets your attention. Not all emails are created equal and they shouldn't be treated as such.

Let's start with what your inbox actually contains and how to sort it.

You've probably got a pile of machine-generated mail that is safe to ignore. Notifications, receipts, platform alerts, newsletters you've been meaning to unsubscribe from. You know the stuff. The sort here is to work out what you should just auto-delete, what to file, and what to pay attention to by exception.

The second pile is mail you need to note but don't owe a reply. Reports, updates, a copy of something for the record. Sometimes you read it. Sometimes you leave it to read later (a time which may or may not come). This is where AI starts to earn its value. Summaries help, but the more useful thing it can do is point you to which of these you should read in full. It can differentiate by who sent it, by what's in it, and by what you've historically spent time on, and it can put the most salient thing in front of you rather than making it something you need to go and find.

This comes with a warning worth being aware of. A summary is not a shorter version of your email, it's a new piece of text generated from it, and two things about how that works need care. It compresses towards what's typical, so the unusual is the first thing at risk of being dropped. And it isn't stable, so the same message summarised twice won't necessarily hand you the same thing. You can do a great deal about the first of those by being explicit. Tell it to flag the person copied in who has never been copied in before and it will, reliably. But you can only name the exceptions you already know to watch for, and the ones that matter are usually the ones you didn't anticipate. What the triage hands back should be treated as a guide, designed to reduce your cognitive load rather than to replace it.

The third pile is mail that needs a reply or an action from you, and it's the only pile worth splitting.

Most of it is transactional. Scheduling, a document request, a confirmation, the coordination email with three people and four constraints between them. What makes these different is that everyone's objective is aligned and the whole thing reduces to those constraints. People aren't really that interested in your reply, they're interested in having the meeting booked. In these cases nothing is lost when you hand it over to a bot, because there was nothing in there to lose. In my setup the AI reads my calendar and handles the drafting, which I approve by hitting send. This is the pile where the direction of travel is fairly obvious: two agents settling a meeting time between them will land somewhere better than two humans and six emails. Or we’ll find a way to skip email completely. Here’s hoping.

The rest is where care should be taken. These are the emails where you're reading signal, not just information.

A number of studies have looked at how AI use changes the way a message is received. Cardon and Coman surveyed 1,100 professionals last year and found that heavier AI assistance made messages read as more professional and made the sender read as less sincere. Others have tested whether it matters if the AI use is declared or merely suspected, and the pattern there is that the penalty attaches to the disclosure and the suspicion rather than to the assistance itself.

Some commentators make a case for using AI to review and polish these emails. Others swear they've never knowingly read an email to the end if they think it was written by AI. Personally, I find AI hugely useful for working out exactly what I'm trying to say, and for testing how it will read to someone coming to it fresh.

The point is that these emails need care, and having AI do an initial sort of your inbox puts you in a better position to give them the time and attention they deserve.

If you're not in a position to connect AI directly to your mail, and plenty of firms aren't, you can still keep the discipline. The sort doesn't live in the tooling. It lives in a rule set you can state, and a connector applies that faster without writing it for you. Which is a decent argument for writing it first regardless.

None of which is really about email. Your inbox already sorts itself, badly, by whatever arrived most recently and whoever was most insistent. All you're doing here is replacing that with a sort you chose. The piles themselves are ordinary. The judgement about what belongs in which one is yours, and it's the part that doesn't transfer from someone else's setup into your inbox. AI makes the sorting easier to personalise and systematise.

This week: take a dozen or so emails from last week that you'd be comfortable pasting into a chat window, spread across the different types you get, and ask an AI to sort them into the three piles with a reason for each. Then argue with it. Where you disagree is a rule you hold and have never written down, so write it down. By the end you'll have half a page that tells you what deserves your attention first, and it works whether or not you ever connect anything to it.