Your first small thing.
You have read the three letters, or someone sent you the link. Either way, here is a doorway.
None of what follows requires an account you do not have, a subscription, or more than sixty seconds. If you already have ChatGPT or Claude open on your phone or your laptop, you have what you need.
One thing to hold, before you start. The point is not to succeed at the task. The point is to notice what happens inside you when you read the reply. Do you trust it. Do you want to push back. Do you want to try again with different framing. That noticing is the muscle you came here to build.
One. Dinner from what is already there.
Open ChatGPT or Claude. Tell it what is actually in your fridge right now. Ask for three dinners you could make this week with what you have.
Here is what I have in my fridge right now: [your list]. Please suggest three dinners I could make this week from these ingredients. Assume I have the usual pantry basics (oil, salt, pepper, pasta, rice, onions, garlic).
What to notice: do any of the suggestions feel like something you would actually eat tonight? Do any of them feel wrong? That contrast is the beginning of calibration.
Two. A thing you keep forgetting to check.
There is something you keep meaning to look at. A package you are waiting on. A bill that comes every few months. The library hold list. A renewal date. Name it to yourself, then ask for the simplest way to remember it.
I want to remember to check [the thing] about once a week. What is the simplest way to do that without adding another app to my life?
What to notice: which of the suggestions you would actually use. That is information about where the agent-shaped hole in your life is, and where it is not.
Three. A draft you have been avoiding.
There is an email you have not written yet. A thank-you note. A hard reply. A decline to something you do not want to do. Pick one.
I need to write a [kind of email]. The context is: [two sentences about the situation]. I want it to sound [warm / direct / professional / casual / something else]. Please draft three versions so I can pick.
What to notice: does any version sound like you? Does any version say something you did not know you wanted to say? Does seeing three versions help you find your own?
One more thing.
What you just did was not, strictly speaking, use an agent. You used a large language model. ChatGPT and Claude are very good at reading a prompt and writing a reply. That is what they do. They do not, on their own, remember you between conversations, check things while you are not looking, or touch anything else in your life.
An agent is built on top of a large language model. It adds three things the model by itself does not have. Memory, so it knows what it has done for you before. Persistence, so it keeps working when you are not there. And tools, so it can actually reach into your life beyond typing back at you.
Now imagine the three things you just tried, as agents instead.
Dinner. Every Sunday evening, without you asking, a short note arrives. Here are three dinners for this week. The calendar looks tight on Tuesday so I picked a quick one. Book club is at your place Thursday, so I picked something that holds overnight. You did not type anything. The agent has been watching the grocery list, the calendar, and the shape of your week.
The thing you keep forgetting. Every Monday morning, one short note. The library has two holds ready. Your passport renews in 127 days. The quarterly bill came in 11% higher than last time, probably the rate change. You do not remember to check, because the agent checks.
The drafts you have been avoiding. Every morning, three short drafts sit in a folder, one for each email you have been putting off. Written in your voice. Approve, edit, or send. You did not write them. You did not even decide which ones to write. The agent saw the threads waiting, drafted the replies, and left them here in case you want them.
You are not there yet. What you did today is the doorway, not the house. But the doorway matters. The muscle you just practiced — noticing how a reply makes you feel — is the same muscle you will use when the agent version arrives. It is the muscle you will use to decide, at every step, which parts of your life the agent gets to touch, and which parts you keep.
That is the journey.
Take care of yourself.
—