The way to do things here is to ask.

They can turn their hand to almost anything you ask. But they are starting completely fresh. They do not yet know how you like your notes kept, which projects matter, who the important people are, or where things should live. So you tell them — naturally, as you go, the way you would brief any new colleague:

That is the whole interface. You ask, you mention, you correct — and 1Presence does the thing, then quietly files away what it learned about you.

So the early days are worth a little generosity. The more you let 1Presence in on how you work — your projects, your people, your preferences — the more it feels like someone who has worked alongside you for years.

Every account starts with one agent, and its name is Presence. It is a generalist — it will do the ordinary work of the day with you, whatever that turns out to be. It is also the one that looks after everything else you set up.

Because you will set up quite a lot. An agent that watches your inbox. A skill that knows how you write a proposal. A routine that runs every Friday evening. A workflow with four stages and a different specialist on each. Presence can read and rewrite all of it, and it can see how their runs went — so there is one place to go when something needs changing, and the same place to go when something came back wrong.

The specialists stay specialists. Your inbox-keeper keeps your inbox; it is not the one you brief about restructuring a workflow. Each of them knows who else you have, so ask one for something that is a sibling’s job and it says so — and offers to hand over, with the message already written. That is why every Edit button in the app opens a conversation with Presence specifically, and why you cannot swap the agent in that conversation for another — overseeing the setup is its job, and it is the one holding the whole picture.

On an agent, a skill, a routine, a dashboard, or any stage of a workflow, you will find a button that says Edit — Edit Routine, Edit Skill, Edit Stage 02. It opens a conversation about that exact thing, with Presence already holding the current version in front of it. Then you say what you want changed.

It reads the real stored instruction before it touches it, so you are always revising the thing that actually runs rather than a reconstruction of it. And when a run has gone wrong, the button on the failure carries the reason across with you — so the conversation starts from what broke, not from a blank page.

A form would be the obvious way to build this. Every one of these things has parts you could put in boxes and a Save button at the bottom. The reason there is no such screen comes down to what is actually inside them.

What you are editing is prose

An agent’s character is not a set of options — it is a few paragraphs of English describing who it is and how it goes about things. A skill is a written method. A routine is one instruction, spelled out in full, handed to an agent afresh each time it runs. A workflow stage is the same instruction problem, four or five times over. Underneath the cards and the timelines and the little coloured avatars, nearly everything you build here is writing.

A form wrapped around a paragraph is a text box with a label on it. It can hold the words. Finding them is the job.

And writing for an agent is its own craft

Instructions that an agent works from are read by a model, and what makes them land is not quite what makes good writing for a person. Order matters more than you would expect. Saying what to do beats listing what to avoid. One worked example does more than three careful sentences. Naming the exact tool for a job saves a wrong turn three steps later. Leaving out the obvious is how a job comes back half-done, because nothing here is obvious to something that was not in the room.

That is the kind of text an agent is unusually good at producing — and it starts with advantages you do not have. It can hold the whole current definition at once. It knows which tools exist, what your other agents already do, and what this one is allowed to reach. It can rewrite the wording against the thing you just said you wanted, then rewrite it again. So the sensible division is the one you would want anyway: you bring the intent — shorter, less formal, stop including the standups — and it does the phrasing.

And it keeps the pieces in step

Most things here come in parts that have to agree with each other. A workflow stage has the full instruction the agent runs, and the one-line summary printed on the card you actually read. Change one of those in a form and the other quietly goes stale — leaving you looking at a card that no longer describes the stage, which is worse than a card with nothing on it. So a rewritten instruction that arrives without a fresh summary is refused outright, and in conversation you never meet the rule: you say what you want the stage to do, and both come back rewritten together.

What stays a button

Plenty here stays a single tap, and should. Run it now, pause the schedule, resume it, delete it, swap which agent handles a stage, add a stage or drag one into a new order — all buttons, exactly where you would expect them. The switches stay switches; the writing gets written.

There are no commands to learn. Say it the way you would say it to a person — full sentences, half sentences, however it comes out. A taste of the range:

Four places to start once the idea has clicked — and if you would rather begin from something already built, a Kit stands up a whole corner of your life at once, keeper, routines and dashboard already wired.