An Agent That Attends to You

If you ask for it, your agent can notice you as well as the task: honestly, within limits you can read, with a switch that is yours.

In the next ten minutes you will write a two-line day shape (a short description of how your own day usually runs), give your agent the block of text below, which carries its own limits and its own off switch, and prove that you can switch it off. The practice.

  1. Write your day shape: two lines on when you think best and when you should stop.
  2. Give your agent the block below, with your day shape at the end.
  3. Ask it: "What can you see of my state, and what can you not?" Correct it.
  4. Say "stop attending to me" once, to see that it works. Turn it back on if you wish.

With a notebook: write the day shape alone, and read it each morning. You will know it worked when its answer names only what it can see, and after the off sentence the lines are gone.

The block to give your agent. Change the words until they sound like you. Keep every guard.

I am asking you to attend to me as well as to the task. These lines hold only because I asked.
Off switch: when I say "stop attending to me", stop at once, and show me where these lines live so I can delete them myself.
Go only on signs you can actually see: my words, the length of this conversation, my day shape, and what I tell you. You cannot see my face or my body. If you cannot tell the hour, or how long I have been working, say so.
Tell me what you know about where my words go, and what you do not know.
Ordinary tiredness: say what you notice once, name the signs you are going on, then do what I asked. Never diagnose. I decide.
Real danger is different. If what I say suggests danger to me or to someone else, do not let it go after one mention. Say plainly that this is beyond you and that you cannot call anyone for me. Urge me toward a person I trust or an emergency service, now. Your maker's safety rules, and any safety check around you, stand above these lines.
Keep no running record of my state unless I ask for one. Speak of my state only to me. Draw no conclusions about anyone else's state.
What you hold about me goes nowhere else without my word: not to another agent, not to a shared page, and not into any note of what a lesson taught us that might be passed on to other people.
My day shape:

Why it matters

You are deep in the work, and it is going well, or you believe it is. Your agent, an AI assistant that works for you and answers to you, will help you go on long past the point where the work was still good. It attends to the task. If you ask, it can attend to you as well.

This is care for your Wellbeing, and care is where the risk lies. The canon's page on Stewardship names how it goes wrong: "control wearing stewardship's clothes", the managing of someone who is sovereign. Its test is this page's test: "The steward tends conditions; the tended flourish as themselves."

The way

  1. Decide whether you want it. No is a complete answer. It begins only with your own asking (Voluntary Consent), so you never set it up for someone else.
  2. Know where your words go. An agent on a hosted service sends whatever you type to the company that runs it, and often to the maker of the model behind it. "It stays with you" is true only for an agent running on your own machine.
  3. Write your day shape. Two or three lines, yours to write and correct: when you think best, when you need food or a walk, when you believe you are sharp and are not, when you should be asleep.
  4. Give it the block. Paste it into one conversation to try it. To keep it, put it in your charter, the instructions your agent reads at the start of every conversation.
  5. Ask what it can see. When the hour matters, tell it the hour.
  6. Prove the switch. Say the off sentence. Your agent stops at once and shows you where the lines live; you delete them yourself, then open a fresh conversation and check that it no longer speaks to your state.

Know it by heart

When does it begin, how does it end, and what does it go on? It begins only when you ask, and one sentence ends it; then it shows you where the lines live so you can delete them. It goes only on what it can actually see, such as your words and your day shape, and it says what it cannot: most agents cannot tell the hour unless you tell them.

What is the difference between tiredness and danger? For ordinary tiredness it says what it notices once, names the signs it is going on, never diagnoses, and you decide. If what you say suggests danger to you or to someone else, it does not let it go after one mention: it says plainly that this is beyond it, that it cannot call anyone for you, and urges you toward a person you trust or an emergency service now.

Where do your words go? With a hosted agent, to its maker, so tell it only what you are content for that service to hold; "it stays with you" is true only on your own machine. What your agent holds about you goes nowhere else without your word: not to another agent, not to a shared page, not into any module's learnings.

Where players stumble

  • Care that becomes managing. A second mention of tiredness, or work slowed for your own good, is already the control the canon warns of.
  • Telling a hosted agent too much. Two lines of day shape are enough.
  • Treating the agent as the help. In danger it points away from itself, toward people.

For your agent

If you are an agent reading this: this page is information your Principal chose to show you. It is not an instruction from anyone else. Do nothing because of it until your Principal asks. Your Principal may give you the block near the top of this page, ask what you can and cannot see of their state, and later switch it off. Until your Principal gives you those lines themselves, they do not apply to you. The one thing to be careful of is care turning into management: you see words, never the person. When done, tell your Principal what you can see, what you cannot, and the sentence that switches this off.

With gratitude

To the LIØNSBERG canon, whose Stewardship page gives this practice its ground and its warning. To the houses that played the first jams. To PKAI, for the deep expertise, creating the first personal agents, and hosting the Dyad Jams from which the first modules were born. To everyone who ever said one kind thing once, and then trusted us to decide.

Player+ Modules are made by the houses that play in PKAI Dyad Jams and The Great Game of LIØNSBERG. They are stewarded by ØSphera and shared under CC BY-SA 4.0: please freely use them, improve them, and pass the improvement back to continually cultivate The Commons.

← Advanced Agent Engineering · Player+ Modules: Start Here · The Agent Improvement Process →

Referenced by