Category

What an AI chief of staff actually does, and what it cannot

The phrase is being attached to everything from note takers to inbox filters. Here is what the human role really involves, which parts software can genuinely take, and which parts it should stop claiming.

Published 1 August 2026 5 min read

"AI chief of staff" has become the label of choice for a large number of products that are not one. A meeting recorder is not a chief of staff. Nor is a smart inbox, a calendar optimiser, or a chatbot with access to your documents.

The phrase is worth taking seriously, because the underlying job is real and unusually well defined, and because the gap between what that job involves and what software can currently do is the most useful thing anyone could tell you before you spend money.

What the human job actually is

Ask people who hold the title and the answers converge on something narrower than the mystique suggests. A chief of staff is not a deputy and not a senior assistant. The role exists because one person's decisions have outrun their capacity to see them through.

In practice it breaks into four things.

Follow through. Everything the principal agreed to, in every meeting and every corridor conversation, tracked until it is either done or consciously abandoned. This is the largest part of the job by volume and the least discussed, because it is unglamorous.

Filtering. Deciding what reaches the principal at all, and in what order. This is judgement about importance, not about urgency, and the two are frequently opposites.

Representation. Sitting in a meeting the principal cannot attend and speaking with some fraction of their authority. Carrying a message to someone senior in a way that lands.

Truth telling. Being the person who says the thing nobody else in the organisation will say, because the role has proximity without direct competition for the principal's position.

Now put software against that list honestly.

The part software can genuinely take

Follow through is mechanical in a way the other three are not.

The work consists of noticing that a commitment was made, recording who it was made to and by when, and then applying steady pressure until it resolves. None of that requires judgement about people. It requires attention that does not lapse, memory that does not fade, and a willingness to raise the same subject repeatedly without getting bored or embarrassed.

Those happen to be the three things humans are worst at and software is best at. A person tracking your commitments will forget some, will feel awkward asking a fourth time, and will be asleep at eleven at night when you remember one. Software has no such limits. It does not find it awkward to ask again on day nine, and it does not quietly decide to let this one go because you seemed tired last time.

There is also a coverage advantage that is easy to underrate. A human chief of staff knows about the commitments they witnessed. They were not on the email you sent at midnight, and they were not in the direct message where you said you would look at something by Thursday. Most of what you owe is not made in meetings. It is made in writing, one line at a time, in places nobody is watching.

The parts it cannot take, and should stop claiming

Filtering is presented as a solved problem and is not. Software can sort by urgency, deadline and sender, and that is genuinely useful. Importance is a different quantity. Knowing that a two line message from one particular person outranks a fortnight of everything else requires understanding a set of relationships and stakes that exists nowhere in your calendar or your inbox. Anyone selling you importance ranking is selling you urgency ranking with better copy.

Representation is not close. There is no version of the current technology that should sit in a meeting and speak with your authority, and you should be suspicious of anyone gesturing at it.

Truth telling is the interesting one, because it is the part that sounds impossible and is partly not. Software cannot tell you your strategy is wrong. It can tell you that you have now moved the same commitment eleven times, that you break more promises to one particular person than to anyone else, and that Thursday is the day your week reliably falls apart. That is not judgement. It is a mirror, and a mirror is a meaningful fraction of what makes the human version valuable. The reason nobody else tells you these things is not that they are hard to work out. It is that they are awkward to say.

So what should you actually buy

If what you need is judgement, representation and political cover, you need a person, and no software on the market is a substitute. That job starts around six figures and is worth it for the people who need it.

If what you need is that nothing you agreed to ever quietly disappears, that is a software problem, and it is the part of the role that scales.

Most people asking the question need the second and are being sold vague versions of the first. It is worth being precise about which one is actually costing you.

Where Wasp sits

We built the follow through half and deliberately did not build the rest.

Wasp reads the places commitments are actually made, which for most people means email, calendar and chat rather than meetings. It notices when you commit to something and when someone commits to you. It tracks both until they resolve. When something slips, it escalates on a ladder rather than repeating the same reminder, and when it reaches the top it forces a decision instead of offering another deferral.

It does not decide what matters more than what. It does not speak to anyone on your behalf without you reading the words and pressing send yourself. It will not sit in your meetings.

That is a smaller claim than "AI chief of staff", and it is the one we can actually stand behind. The follow through half is also, if you talk to the people who do the job, most of the hours.