When AI Helps With Little Real-Life Things
AI is everywhere, and people normally think of it as something that can help with big endeavors: creating apps, make us rich, world domination, etc. But after two years of using it myself for all kinds of things, I find it specially useful when applied to the daily boring little things.
I am going to describe the most recent example, which started with a very normal dad problem: I was setting up a laptop for my 9-year-old, and I wanted the boring, reasonable things parents want. No random internet, no surprise YouTube rabbit holes, no scary search results, and some way to help him when he got stuck.
I rescued an old PC and installed Linux Mint, knowing quite well that parental control was going to be an issue. No Fa,ily Link app like on his Android tablet. But the point is that I want him to experiment with programming and also use Canva, so I was determined to find solutions.
Yes, on phones and tablets, you can reach for Family Link or Screen Time and control absolutely everything. On a Linux laptop, the parental-control story is not as polished, and it’s actually something that got me thinking about products (but not today!). There are good pieces, but they are scattered: DNS filtering, browser policies, login limits, remote desktop, system services, shell scripts, logs. Useful parts, not one friendly parent button.
I have used Codex to examine the laptop and all the places I could hook into, to make my kid’s experience safer. Turns out there’s a lot you can do, but where it ended up, I did not see that coming!
The first version was just safety
The first job was web access, that’s the biggest thing and I am surprised at how limited control parents have on Linux, unless they are techies and are ready to manually edit a bunch of files. Linux can’t become mainstream without this, it’s an absolute must-have.
I wanted a walled garden, not a blacklist. Email, Canva, Scratch, Wikipedia. That kind of thing. If my kid needed YouTube, he could use his tablet, where the parental controls were already better suited to that job.
I started tinkering with Codex, and we set up DNS filtering with NextDNS and locked the browsers down with allowlist policies. The laptop could resolve and open the sites I approved. Everything else should bounce.
That sounds simple until a child tries to use the thing. Gmail is not just Gmail. Login runs through Google accounts. Some flows touch account management. Canva and Scratch pull assets from multiple hosts. A kid does not say, “Dad, please approve this clean root domain with no subresource surprises.” He says, “It doesn’t work.”
The setup had to become observable, so blocked attempts went into a log. The allowlist became editable through a controlled command. The browser policies could be regenerated. The browser could be closed remotely so policy changes took effect.
At that point I had a safer laptop. What I did not have was a parent-friendly control surface.
Then Hermes got a job
Hermes is my agent running on a small local machine at home, which also runs AI models fully local, so it’s all private and safe. It already talks to me through Telegram, and the laptop now had enough safe controls for Hermes to become the parent console.
The dangerous answer would be to give Hermes SSH access and let it do whatever it wants. I did not want that, so I enabled a narrow command channel. Hermes connects to the laptop with a restricted key, but it never gets a shell.
The command channel can still do useful things. It can show laptop status, list or approve blocked websites, open an approved URL, close the browser, start the help app, send my son a desktop notification, set the wallpaper, read battery health, and manage today’s computer time. Locking, shutting down, and rebooting sit behind confirmation.
That distinction matters a lot. “AI has root on my kid’s laptop” is a terrible sentence. “AI can press these labeled buttons, and some buttons require confirmation” is a design I can sleep with. Hermes is now a shortcut. I can still SSH into the machine, but it’s way faster to just use natural language and ask Hermes to do things for me.
The Ask Dad app made it feel human
My son does not have a phone, so if I am away he’s got no way of asking me how to do something if he gets stuck. So I created a tunnel app called “Ask Dad”, where my son can ask questions. I get them on my phone (another Telegram bot) and I can answer directly. He can even send me screenshots of what he’s doing just by pressing a button. If he is stuck, he does not need to describe the whole screen from memory. He can show me.
That sounds small, but it makes the experience less intimidating for him, it’s his first laptop ever! I didn’t want him to to see remote help as an adult workflow: open remote desktop, call me, explain the problem, wait while I connect. Now it is a kid workflow: write “no funciona”, tick screenshot, send.
We added the little things that make it feel less like a prototype. Only one instance can run, so Telegram polling does not fight itself. Enter sends the message. Sending does not pop a modal that steals focus. Screenshots hide the Ask Dad window first. The history can be cleared from the setup screen, behind my PIN. Hermes can reply into the app, but my kid is not chatting directly with an unrestricted AI.
That last bit matters to me. I like agents, but I surely do not want my 9-year-old casually negotiating with one (yet).
Ask Dad gives us a safer pattern. He talks to me. If I want Hermes to help draft, translate, summarize, or explain something in kid-friendly Spanish, I can route the message through Hermes with a parent-controlled prompt.
Each hook exposed the next chore
Once Hermes had a safe command path, ideas arrived faster than the setup could complain. A read-only inventory command answered “what apps are installed?” Educational launches went through a fixed approved list. Wallpaper changes became a small pipeline: stream the image into a command, validate the MIME type, store it in my kid’s pictures folder, and set the desktop wallpaper.
URL approval needed more care. Hermes can inspect the URL, decide whether an exact URL or full domain makes sense, update the allowlist, refresh browser policy, then open it for my kid. Exact URL approval is useful when the page looks fine but the whole domain does not deserve trust.
The laptop status became one Telegram command: battery, Wi-Fi signal, CPU, memory, disk, services, Ask Dad status, browser count, and pending requests.
Time limits needed a real tool behind them. We used Timekpr-next, which already understands session limits. Hermes can now say “30 minutes left”, “no time limit today”, or “time’s up today”. The lockout mode terminates my son’s session instead of merely locking it, because my son knows his password.
And because kids should see the rule, not just suffer the rule, we added it to the desktop widget.
The desktop became a little dashboard
The laptop now has a small Conky widget on top of the wallpaper. It shows the weather, remaining PC time, Wi-Fi status, battery, CPU, memory, disk, Ask Dad status, and whether protection is active.
This is partly useful and partly emotional. Kids should be able to see the rules around them.
If the kid has 20 minutes left, he can see that. If the battery is low, he can see that. If Ask Dad is ready, he can see that too.
The computer stops being a black box with rules hidden inside it. It explains itself a little, which is quite cool.
We even added a dedicated Hermes command for battery health using UPower, because “the laptop is at 94%” is not the same as “the battery has 74% of its original design capacity left after 346 cycles.” I do not need that every day, but when I do, I want it from my phone.
The point of “AI assisted parental control”
The point was to remove friction from the parts of parenting that are, unfortunately, systems administration.
Approving a website is not a moral essay. It is a small decision followed by five annoying technical steps. Closing a browser after a policy change is not parenting, that is simply plumbing. Checking whether the safety service is running is not quality time. Hermes handles that plumbing.
I still decide what gets approved. I still decide when time is up. I still decide whether a reply should go to my son. The agent is useful because it makes those decisions easy to execute from Telegram while I am in the kitchen, on the sofa, or away from the laptop.
The funny part is that none of the pieces are magical. NextDNS filters DNS. Browser policies enforce the garden. Timekpr manages session time. Systemd runs services. UPower reports battery details. Conky draws desktop text. Telegram moves messages. SSH carries commands. Linux does Linux things.
AI made those pieces feel like one system, and that is the part I keep thinking about.
This is the kind of agentic software I actually want
I do not want agents wandering around my home network improvising. Give me agents that know exactly which buttons they are allowed to press. This little parenting project pushed me toward narrower access, made useful, instead of the usual agent hype about expanding autonomy.
A good agent command should have a name, an input shape, a permission boundary, and a human expectation attached to it. battery-health is read-only, so it is safe. set-time-left 30 changes my son’s day, but the action is bounded. time-up-today is powerful enough that Hermes should ask me first. allow-and-open-url is convenient, but it should inspect the URL and think about scope before touching the allowlist.
This is where AI can help with real life: it sits between intent and operations, while judgment stays with the parent. If I say, “Kid has 30 more minutes,” Hermes turns that into the right Timekpr command, refreshes the desktop widget, and the laptop enforces it. If I ask what my son tried to open, Hermes reads the blocked-attempt log and summarizes it. If I say, “Tell my son I will help him in five minutes,” Hermes sends a desktop notification or an Ask Dad reply in Spanish.
I love that this actually looks like science fiction, but it is boring dad ops behind a cool workflow.
What I would do differently next
This whole thing grew in a few hours, which is both the charm and the warning label.
The shape is good, and with a bit of hardening it could be a product. Every command should be logged. Secrets should stay out of configs that kids can read. Destructive commands need confirmation, broad domain approvals should be avoided when exact URLs are enough, and the laptop needs a recovery path for the days when the agent machine is offline. I would also document every command in the Hermes skill, with examples.
I would also keep the system boring, which sounds like an insult but is the highest compliment here. Boring means the laptop is not dependent on a fragile chat trick. If Hermes goes down, Timekpr still enforces time. NextDNS still filters. The browser policies still exist. Ask Dad still has local files. The agent is the nice interface, not the thing holding the wall up.
A few hours later
We started with “Linux parental controls are not quite enough for what I want.” A few hours later, we had a local-first parent console with a walled garden for approved sites, a Telegram-accessible agent, a kid help app with screenshots, educational app shortcuts, desktop widgets, time limits, URL approval, remote notifications, laptop health checks, and wallpaper changes, because delight matters too.
I did not plan the whole system up front. We hooked it up piece by piece, following the next annoying thing, which is probably why it worked.
Most useful AI systems in daily life will not arrive as grand platforms. They will start with one irritating gap between “I know what I want” and “the computer makes me do twelve steps to get it.” Close that gap carefully, then close the next one.