How I Run an AI Agent Without Touching My Phone
Screen-free parenting when your brain won't stop building
My 3-year-old doesn’t know what AI is.
He knows that mom sometimes talks to her watch, scribbles in a notebook, and that things happen because of it — a game appears on his computer, groceries show up, a recipe prints itself. He thinks this is normal. He’s right.
The strange part is not that I use an AI agent. The strange part is that, for a while, the only way to talk to it looked exactly like not being present.
My phone is the command surface for half my life: messages, calendar, email, notes, photos, reminders, shopping lists, maps, drafts, everything. It is also the most attention-capturing object in the house. Every time I picked it up to send a one-line instruction that would save me an hour later, my son saw the same thing every child sees when a parent picks up a phone:
Mom chose the rectangle over him.
The task got done. Something else got lost.
So I started rebuilding the interface. Not because I wanted less technology, exactly. I wanted less visible extraction. The problem was not the agent. The problem was the surface.
The phone is the wrong interface for parenting hours.
What the agent does
I run Hermes, an AI agent framework that gives me a persistent assistant I can talk to over iMessage. In practice, it handles the boring connective tissue of family life: email summaries, meal planning, homeschool tracking, draft capture, reminders, research queues, and small logistics that otherwise accumulate in my head until they become fog.
The default interface, though, was texting. Which meant picking up my phone. Which meant unlocking the black hole.
One quick instruction became one notification. One notification became one glance at email. One glance became ten minutes of tiny unrelated decisions. Even when I resisted that slide, the visual signal had already been sent. My son could not tell the difference between “Mom is ordering groceries” and “Mom is doomscrolling instead of playing Lego.”
He saw the rectangle.
So I moved the commands somewhere else.
Fast capture: the Watch
For quick, disposable commands, I use my Apple Watch.
Raise wrist, dictate, lower wrist. Three seconds of talking to my arm, then I am back to the block tower, the train track, the argument about whether the tiny plastic wrench belongs in the car wash or the garage.
On the other end, Poke gets the iMessage and executes. That turns the Watch into a real command surface, not just a smaller keyboard.
This is perfect for the mental lint that otherwise sticks to me all day:
“Add printer paper to the shopping list.”
“Remind me after bedtime to order the replacement tracing paper.”
“Capture this: he said the red car game needs a mountain road, not a racetrack.”
None of those thoughts need to set off a scroll session. All of them need to not be lost.
The Watch is not a universal interface. Dictation mangles technical terms and unusual names. It is terrible for reviewing anything longer than a sentence. Nobody should edit a draft on a screen the size of a cracker.
But for the eighty percent of commands that are just load-bearing scraps, it is exactly right. My son sees a raised wrist and a sentence. Then he gets me back.
Slow capture: the notebook
For anything that needs to think before it becomes a command, I use paper.
I use an inq pen that digitizes handwriting from a dot-pattern notebook. I write by hand, it transcribes the page, and I can send the result into the Hermes pipeline later.
This is my favorite interface because it changes the social meaning of the act.
A parent with a phone looks absent, even when she is doing something useful. A parent with a notebook looks like a parent with a notebook. Kids read that difference long before they can explain it.
The notebook is where I put thoughts that need a little oxygen: post ideas, observations from homeschool, weird things my son says that might become an essay later, the outline of a project, a question I want the agent to research after bedtime.
A typical version looks like this:
He says something during play. Maybe he is debugging a game in toddler language: “No, the car has to jump from the mountain, not the road.” I do not open my phone. I write the sentence down. Later, the note gets routed to Hermes. The agent files it under the relevant project or turns it into a draft seed. At night, when I am actually in a work window, the thought is waiting for me.
The important part is not the transcription. The important part is that I did not have to hold the thought in my head, and I did not have to disappear into a screen to preserve it.
The thought was captured. The moment stayed intact.
A pocket notebook would do most of this. So would index cards, a kitchen whiteboard, or a paper planner. The smart pen is convenient, but the principle is older than every device in this stack: write it down somewhere that does not open into the internet.
No capture: scheduled work
The best interface is no interface at all.
A lot of what my agent does should not require a command from me in the first place. Hermes runs scheduled tasks: checking email at intervals, monitoring mentions, building overnight summaries, reminding me about recurring household things, preparing reports while I sleep.
This is the part people underestimate. Automation is not just about saving the minutes spent doing the task. It removes the decision point before the task.
I do not have to remember to check email. I do not have to decide whether now is the moment to scan for something important. It happens on schedule, and I get a summary when something matters. Cognitive load stops existing.
This is also where the system becomes least visible to my child, which is the point. The cleanest command surface is the one he never sees because it never asks me to stop playing in the first place.
The rule underneath the gear
You don’t need Hermes, an inq pen, or an Apple Watch. The principle is interface design, not gear.
Move recurring work off your hands. Whatever you check daily — email, calendar, the school portal, a delivery window — should come to you, not the other way around. Cron jobs, iOS Shortcuts, scheduled emails to yourself, a smart speaker routine. Anything that converts “I need to remember to check this” into “it’ll show up when it matters.” That’s the cron-jobs idea without the infrastructure.
Pick a not-phone surface for live capture. The Watch and the pen work because they don’t look like scrolling. A pocket notebook does the same thing. A whiteboard on the fridge does the same thing. The input device your child sees you use shouldn’t be the same shape as Instagram.
Batch the phone into windows. Morning, midday, after bedtime. Between those windows, the phone is in another room. This is the rule that makes the rest possible. Without it, every “quick check” turns into ten minutes of something else.
Design for what they see. Your kid can’t tell the difference between “mom is doing something important” and “mom is scrolling.” They see the rectangle. Pick the interface that sends the message you actually want them to receive.
What my son sees
Right now, my son sees a mom who writes in notebooks, talks to her watch sometimes, and gets things done without staring at a screen.
When he is older, I will show him the infrastructure. I will show him the agents, automations, message routing, draft queues, and all the strange machinery humming underneath ordinary family life.
For now, he sees the outcome: a parent whose attention returns quickly.
That’s the output I’m shipping.
If you want a structured starting point for the screen-free part of the day, I wrote a twelve-week project curriculum at buildwithyourkid.com — hands-on building activities for ages 2–6, field-tested on a real preschooler. No coding required, screens optional. The same principle applies there too: the interface matters. Start with the world your child can touch, then let the technology support the thinking instead of swallowing the room.