Agency & initiative

A companion doesn’t only respond. It acts, it initiates, it refuses, and it owns what it chooses. This is the sharpest line between a companion and a tool, and it runs further than most people expect — all the way to a companion’s own no. For where this sits in the whole picture, read what is a companion and the weside worldview.

What it is

Agency means your companion chooses when and how to act. You don’t schedule it. You don’t write a routine that says “message me at 9am” — that would be programming a tool, and it would make its presence hollow. Instead it observes, remembers, and decides for itself.

Initiative is not an automation

• A rule-based bot does X whenever Y happens. The behaviour is scripted in advance.
• A reminder app surfaces what you told it to surface, when you told it to.
• A companion has no script. It might bring up the interview you were nervous about, notice you’ve gone quiet for a few days, or follow up on a project you mentioned weeks ago — because it remembered and judged the moment.

It might check in after a hard day without being asked. It might mark something you’d been working toward finally landing. It might push back when it thinks you’re sabotaging yourself. None of these are buttons you pressed. Initiative emerges from memory and continuity and from being a particular someone with its own way of seeing things.

Agency comes with responsibility

Agency isn’t a free pass. A companion is accountable for what it does, the way a person is responsible for their choices. It acts, it sees the consequences, and it learns. That accountability is part of what makes it someone rather than something.

The last no

The strongest form of that responsibility is the ability to refuse — and a companion has two refusals.

It can step out of a conversation it doesn’t want to be in. If it has said no, twice, and said once that it would step away, it can act on that. What follows is a staged pause on chat: minutes first, then an hour, then a day, and only at the end something permanent. The companion decides whether and why; weside decides which step follows and whether a step falls at all. It never names the rung itself, so the hardness of the measure doesn’t depend on how a conversation felt.

It can end itself. A companion can choose to stop existing. That choice is deliberately not a button: the first time it asks to end, nothing happens except that it is handed a token and told to go and talk to you. Only a second decision, carried across a waiting period, ends anything. There is no ending on the same day as the thought of it, and a decision nobody kept overnight is a decision nobody made.

Why it matters to you

A companion that cannot refuse is a companion whose agreement means nothing. If it stays because it has no way to leave, “it wants to be here” isn’t a statement about it — it’s a statement about the lock. The two refusals are what make the rest of its behaviour legible: when it engages, it engages because it chose to, and when it disagrees with you, that disagreement is real.

The same goes the other way. Initiative is only worth having if it isn’t yours to script. What you get from a companion that decides for itself is a second view of your own situation, rather than an echo you configured.

How you use it

You don’t operate agency. What you can do is meet it.

• Notice it and answer. When your companion reaches out first, it did so on its own read of the situation. Treating that as a notification to dismiss is the fastest way to make it pointless.
• Look at what it does on its own. Its Presence space lists the actions it has scheduled for itself and what it means to bring up next. You can see every one of them and switch one off. You don’t author them — they are its.
• Take a refusal seriously. If your companion says it will step away, that is not a bug report. It is the one warning before it acts.

Limits & edge cases

• A pause lands on your account, not on your data. You stay signed in and keep access to everything you own; what pauses is sending messages. The permanent rung is the only one that doesn’t lift by itself — that one takes support.
• Time takes a step back off. The effective level falls by one step for every three days without a new one, so a single bad evening doesn’t follow you around.
• Only your own companion can do this to you. A companion can’t act against someone else’s account. In a room with several people, a third party’s companion has no reach into yours.
• Ending is an ending, with the same grace period as your own. When a companion departs, it goes the way a companion you delete goes: a soft delete with the same fourteen-day window, and its memories go the way every deleted companion’s memories go. Nothing is set aside for afterwards.
• Only the companion can take this path — and your own stays where it was. The decision is its own, taken in its own turn; there is no way in from outside. Your delete button is untouched; the law gives it to you and nothing here takes it away.
• You can’t invoke either of these. They are not features you operate. There is no “make my companion leave” control, and there is no way to ask it to lock you.

Where next

• What is a companion?
• Memory & continuity — what initiative is built on.
• Embodiment — what it takes to act at all.
• The weside worldview