# Let their assistant ask yours

JP LeBlanc, 2026-09-27. https://humblekind.co/blog/let-their-assistant-ask-yours

> Your assistant can read your calendar. A stranger's assistant asking you something and getting an answer you control is the half nobody built. It works now.

There is an asymmetry in every AI product shipped in the last two years, and once you see it you cannot stop seeing it.

Everything answers the same question: what can my assistant read on my behalf. Sign in with this, connect that, add this server, tick these boxes. It is a solved problem with a standard and a hundred implementations.

Nobody built the other direction. What can somebody else read of me, through rules I set, without me sitting in a chat answering it live. That is the half this is for, and it is now the least demoed and most interesting thing here.

## What it looks like from the asking side

Somebody has a call with you on Thursday. They ask their assistant to find out how you like to work before it starts.

Today that assistant does what a person would do badly: it reads your profile, finds three posts, and produces a confident summary of a person who does not exist, built out of the shape of people who look like you.

With a page, it asks. It gets sentences you wrote, quoted, with a note of where each one came from. If it asks something you never wrote down, it is told that plainly rather than being handed something plausible. And whatever it asks, it gets exactly what its person would get standing at your page themselves, because what it may see is decided before it is allowed to ask anything at all.

The assistant does not have special standing. It is its person, with less patience.

## What it looks like from your side

Nothing, most of the time, which is the point. Somebody's assistant asked, your page answered, and it is in your own log with the question, the answer, and what it drew on.

There is nothing to set up for this beyond the page itself. If your page is open to the web, assistants can ask it the way people can. If you want a specific person's assistant to get more, you invite that person, and their assistant inherits it because it is asking as them.

## And your own assistant, on your side of it

The same connection runs the other way, and this is how most pages actually get written.

You connect your own assistant, the one that already knows what you have been saying all week in documents and chats. It can look at what your page holds, notice you have explained the same thing four times, and propose it as something to add. It shows you the exact words that would be stored. Nothing lands until you say yes to those words.

This is the part that makes a page maintainable rather than a project. Writing down what you know is a chore nobody does twice. Confirming a sentence somebody drafted from what you already said is a thing you will do while waiting for a build.

It works with Claude and with ChatGPT today, and with anything else that speaks the same protocol. Four sentences of setup, which are on[the connect page](https://humblekind.co/connect).

## The rule that makes it safe to plug in

An assistant you connect can add to your page. It cannot change who may see anything. It has no access to your audiences, your invites, or your settings, by construction rather than by policy: those tools do not exist on the connection it is holding.

That separation is deliberate and it is the reason I am comfortable telling people to connect something. The worst case with a badly behaved assistant is a proposal you decline. It cannot open a door.

The same goes the other way. A connection that lets an assistant ask your page as its person is bound to that one page and nothing else. Point it somewhere else and it is refused.

## Why this is worth doing before it is normal

At some point, being asked about by software will be the majority of how people encounter you professionally. It is already a meaningful slice, and it is the slice where you currently have no say at all: something reads whatever it can find and produces a person.

The cheap version of taking control is to write more posts and hope. The version that actually works is to have an address that answers, in your words, with the parts you chose, and let the machines use it.

[Connect an assistant](https://humblekind.co/connect), or[ask a page](https://humblekind.co/explore)with yours and see what comes back.

## More from the blog

- [When someone looks you up](https://humblekind.co/blog/when-someone-looks-you-up.md)
- [Money and health are not in the product](https://humblekind.co/blog/money-and-health-are-not-in-the-product.md)
- [The things we decided not to build](https://humblekind.co/blog/the-things-we-decided-not-to-build.md)