I have grown to hate the smartphone.
It's a glorious piece of tech. But it's a trap built around usefulness.
At 2 AM, I wake up and reach for my phone. I know it's not healthy but I do it out of necessity. I work with people in different timezones. Overnight, decisions get made, messages arrive, deadlines shift. If I don’t check, I risk missing something that matters.
So I check: Email. WhatsApp. Slack. iMessage. LinkedIn.
Somewhere between those apps, the boundary breaks. I find myself no longer checking for important things but consuming whatever shows up next. By the time I stop, I’ve lost time and some tokens of my attention.
This is the real problem with the smartphone:
==The important messages arrive through the same channel as the distractions.==
Your client’s message, your mother’s greeting, and a notification from a social media app all compete in the same space, with the same priority. Unfortunately, the apps that win are the ones engineered to interrupt, not the ones that matter.
You check your phone because the environment you are operating on is designed that way. You are not addicted to your phone. Checking is the only reliable way to not miss something important. Everything else follows from that.
For the first time I broke that cycle. I woke up at 2 AM and didn't touch my phone.
A few days earlier, I built a Claude skill called tomorrow.
Before bed, it pulls my Google Calendar, scans emails from the last 36 hours for anything that requires action the next day, and checks Slack and iMessage for messages that signal requests, deadlines, or decisions. Everything runs through a set of rules I defined.
The output is a clean and prioritized rundown of the next day. It is designed to be easy to read and scan through.
I run it before sleeping. When I wake up, I already know what my day is going to look like. I ran another skill today just to be sure after all my morning preps.
There was no reason to open all those other apps on the phone. I made coffee and started working.
I solved part of the "alert fatigue" problem in my first week with Claude Skills. 1 command doing the work of 6 apps.
From what I can see, we can save ourselves from mindless smartphone consumption once we start applying AI skills as filters and guardians for our attention. This idea isn't new. It's been discussed in research and is currently being applied by other builders across multiple fields. This is my account of how it played out for me.
What skills actually change
A Claude skill is just a folder with instructions and code. You ask Claude to do something, it loads the relevant skill, runs it, and gives you an answer. Nothing magical about the technology.
What matters is the direction of the interaction. With a phone, information comes to you and you sort it. With a skill, you go to the information and it comes back already sorted. The phone is push. The skill is pull.
My tomorrow skill replaces the morning email, calendar, Slack, and iMessage check. I built today, a midday rundown so I don’t open Slack just to see what I’ve missed. deadline handles scheduling without opening Google Calendar or the reminders app. Next is a notification filter that surfaces only actionable items from LinkedIn and email and drops the rest. After that, scheduled email drafting and more complex tasks.
Each skill replaces one reason to pick up the phone.
However, I still need the phone for music, messaging and for the camera. But the involuntary uses, the checks, the just-in-cases will start to fall away. Picking up the phone will eventually become a deliberate act. Something I do for pleasure because the important things have already been handled.
The real shift here isn't necessarily killing the smartphone but cutting the relationship where it owns your attention by default and you fight to get it back.
Why this is different from "just use less screen time or use different notifications."
I've tried screen-time apps. Greyscale mode. App blockers etc.
They all fail because they treat the symptom which is checking your phone frequently. You check apps on your phone because it's the only way to not miss things. The impulse to open and check the phone will go away the moment we remove the uncertainty around the alerts and notifications they bring. Good systems make the right behaviour cheap and the wrong behaviour expensive.
...But you still need your phone don't you? ...and other gotchas
Yes.
Skills run on a phone too.
I can trigger tomorrow and deadline skills from the Claude app on my Android. I'm not escaping the phone. I'm changing what I do on it. There is the phone-as-tool and the phone-as-an-attention-sinkhole . These are two different things that share the same hardware.
Skills are limited because they only work if you build them with the right context for your needs. Most people won't. The barrier is somewhere between users who know what a markdown file is and users who can write basic instructions. There is no coding involved for entry level skills, but a barrier does exist. Until that barrier drops, skills will be a tool for people already inclined to build their way out of problems.
Secondly, the model could change. APIs break. Permissions change. Anthropic could deprecate skills. The economics on AI could shift (tokens are already getting expensive). I'm not betting on this specific implementation.
Another flaw is the issue latency vs urgency tradeoffs. If you decide to get rid of real time notifications for a sweeping 3 hour skill or agent run, there is a possibility of seeing a notification 2 hours too late.
I'm betting on the pattern of intentional pull-based AI tools that replace specific reasons to grab a phone. If Claude skills aren't the thing that wins, something shaped like them will.
The Future That IBM Saw in 2020:
We are not yet at the point of deleting apps from our smartphones for good. But if one skill can replace the need to open 6 apps then half the battle is one. History has taught us that convenient tech always wins! Maybe the app can be replaced once trust issues around AI are addressed.
In 2020, researchers at IBM published a paper proposing exactly this kind of system: an intelligent notification framework that sits between you and your apps.
It learns which alerts to issue, which to suppress, and which to batch together for later. Their framework classified alerts by severity, modelled user behaviour, preferences and availability. They called it a "snooze-less" system. The goal was to eliminate the need to snooze or ignore notifications entirely. If the system does its job, every notification you see is one that deserves your attention at that moment.
The paper was ahead of its time. In 2020 the infrastructure to build this didn't exist for individuals. You needed enterprise backends, custom ML pipelines, dedicated integration work. In 2026, a Claude skill connected to Gmail and Calendar via MCP does roughly the same thing from a markdown file on your laptop. The gap between research prototype and personal tool collapsed in six years. There is further research being done on why companies should now build notifications for both users and AI Agents.
The IBM paper also hints at a future that goes further than filtering.
If an AI agent can reliably read, classify, and act on your notifications, you don’t need the app at all. You need the protocol: the underlying rules and interfaces that define how messages are sent, received, and interpreted.
Email already works this way. Slack exposes this through its API. WhatsApp is a bit tricky, and the security model there makes direct access harder, but that’s a separate problem. The point is: the app was always just a frontend for a communication channel. If your AI agent can interface with that channel directly, the app becomes optional. Not today. But the architecture is already pointing there.
That's the future I find most compelling is where AI replaces the entire app layer to talk directly to the systems underneath. We will stop interacting with interfaces designed to capture our attention and start interacting with an agent designed to protect it.
What I'm actually claiming
The notification-driven model of computing is a historical accident, not a permanent condition. It won because attention was the cheapest thing to monetise on a small screen with constant connectivity. It will lose to anything that lets us get signal without noise.
Maybe Claude skills are an early primitive version of that anything. They're not pretty. They require setup. The ecosystem is young. There is a lot of push-and-pull. But they're the first tool I've used that actually changes the math of why I pick up my phone.
With the right tools this is how App Independence can be attained:
Step one: Build a scheduled digest skill that pulls Gmail and Slack, iMessage every few hours and filters for what actually needs me.
Step two: Turn off notifications from those apps on my phone.
Step three: Live with that for two weeks and see what falls through.
Step four: If nothing falls through, stop opening the apps entirely.
Step five: Maybe a month out: delete them.
Thanks for reading
All Claude skills referenced in this article are published and maintained here:
Top comments (0)