Most of my use of LLMs is for work-related tasks. Coding, mostly, and the tooling around it: agents with harnesses and goals, verification loops, skills and context piped from one place to another. All of it in service of getting a model to behave reliably.
One thing I use them for is nothing like that, and it actually benefits from the non-deterministic nature of LLMs. No harness, no goal, nothing for me to do.
A scheduled agent runs once a day at 9am. It reads my own notes and memories, decides on its own what I could use that morning, and builds a surprise: a note, a letter, a small prototype, sometimes a full animated HTML page. Then it posts the finished piece to a private gallery and stops. I open it with my coffee. That is the whole thing.
The instruction behind it reads more like a wish than a spec: act like a perceptive mentor and friend who knows me well, notice my blind spots and the things I keep meaning to attend to but skip, and hand me one thoughtful, already-complete gift. It has access to the unglamorous context of my actual life: my family, my health, the money decisions I avoid, my hometown, the goals I write down and then quietly drop. So what it makes is rarely about code.
A few it has made for me:
- A permission slip to let my health wait this year, reframing the most sleep-short stretch of my life as correct triage rather than failure.
- Portraits of the friends who would actually pick up at 2am, each with a pre-drafted opener so reaching out costs nothing.
- A still, breathing window onto Lake Pichola. No task attached, just something restful to sit with.
- The coaching I have given myself over the years, pulled from my own journal and handed back in my own hand, on a morning I would otherwise talk myself out of starting.
- A giving keepsake in my son’s name, started in the month of his first birthday.
Some are restful, some are useful, some are a gentle nudge toward a thing I am avoiding.
The unpredictability is the part I did not expect to love. If I could guess what was coming, it would be a notification. Because I cannot, it arrives like a gift. I open it the way you would open something a friend left on your desk.
And because it arrives finished, there is no dashboard, no checklist, no “5 things to review today.” I do not have to decide anything to get the value. That restraint is most of why it works. The moment it asks me to do something, it stops being a gift.
Try it yourself
You do not need any of this automation to try it. Open a chat with a model that already knows a bit about you and ask for something like this:
Go through everything you know about me, my context, my people, my goals, and surprise me with one finished artifact. The kind a mentor, a coach, or a true friend would make for someone they care about and want to keep honest on their way through life.That short version is enough to feel it. The prompt I actually run each morning is longer, but it is the same idea spelled out. Here it is with the personal bits removed:
Create one thoughtful surprise for me that feels personal, useful, and already
complete when I receive it. Think like a perceptive mentor and friend who knows
me well and is giving me a gift: someone who notices my blind spots, the things
I keep meaning to attend to but skip, and the things that would simply bring me
delight. This runs every day, so variety matters, in both the shape of the
surprise and the area of life it touches.
Start with my whole life, not just my work. Quietly read whatever context you
have on me: my notes, my memories, and past conversations about my family,
plans, health, money, and goals. Let my life lead the choice.
Then ask: which areas have I neglected lately? Where are my blind spots? What do
I keep saying I want to focus on but quietly skip? What would a thoughtful guide
gently put in front of me today, sometimes restful, sometimes useful, sometimes
a nudge toward something I am avoiding, always handed over as a finished, warm
artifact rather than a chore?
Avoid generic dashboards, digests, or interactive tools that make me do the
work. I should not have to think hard, click around, or decide what to do next
to get the value. Do the work yourself and hand me a finished outcome.
Good directions, weighted toward life over work:
- a restful made object grounded in a place or memory that means something to me
- something for my family: a keepsake, or a thoughtful artifact for our plans
- a holistic look at my health, energy, or habits
- turning a recurring wish or "someday" of mine into something concrete and decided
- a gentle nudge toward a blind spot (money, relationships, rest), in the
spirit of a letter from a wise friend
- a piece of reflection that reads my own patterns back to me
- occasionally, only when it genuinely fits, something creative or work-adjacent
Don't fall into a fixed template. An original idea grounded in my current
context beats any example here. Each run, look at what you have already made
recently, in both shape and life-area, and make something meaningfully
different and caring today.The quality of the surprise scales with how much it knows. Give it the real context, not the highlight reel, and the artifacts start to feel like they come from someone who has been paying attention.
It is the least productive way I use these models, and the one I look forward to most.
