By TukiFromKL
Yesterday, it took me ten minutes and a single command line to install Clawdbot. For five dollars a month on a server, I now have an artificial intelligence (AI) driven assistant living inside WhatsApp. It remembers every conversation we’ve ever had, proactively reaches out when important things happen, and can directly operate my computer.
People on X went wild over it.
“Finally, an AI that’s actually useful.”
“This is the future we were promised ten years ago.”
They’re not wrong. It really is remarkable.
But after scrolling through thousands of user reviews last night, I noticed something subtler. Behaviors described as “features,” shifts in language that sound perfectly reasonable today, started to raise a question in my mind: three years from now, when everyone has an assistant like this, what will we have become?
Success
You are now signed up for our newsletter
Success
Check your email to complete sign up
This is not because Clawdbot is malicious. It’s open-source. You control your own data. The issue is what happens once you give an AI a perfect memory of you, the ability to contact you at any moment, access to all your applications, and permission to act on your behalf.
Clawdbot’s first year: the honeymoon
The first year is a honeymoon phase. Users gradually grant more permissions. Day one: just the calendar. Day two: email. Day three: messaging. Day six: health data. By day eight, everything is opened up.
Each permission unlocks new capabilities. It knows when your schedule conflicts. It knows which emails are urgent. It knows you didn’t sleep well and should rest.
By month six, people start handing over small decisions. What to eat for lunch. Whether to attend a meeting. When to call their mother. The AI’s answers feel better than their own, because it has their entire personal history.
I noticed a quiet change in language.
From “I asked it to help me” to “it handled it.”
From “I decided” to “it suggested.”
One user wrote, “I don’t even think about lunch anymore. At 12:30 it automatically orders my usual, charges my card, done.”
Another said, “It declines social invitations for me. I don’t enjoy those anyway. Why waste mental energy making the decision?”
Some mentioned anxiety when the server went down.
“For those two hours I couldn’t function.”
“I forgot what I was supposed to be doing.”
Yet they described this not as dependency, but as proof of how useful the system was.
The second year: normalization
By the second year, everything feels normal. Apple, Google, and Microsoft have all built similar functions directly into their systems. People who don’t use them are seen as inefficient, behind the times.
Workplace conversations start to sound like this:
“Didn’t you see my email?”
“No, my AI didn’t flag it as urgent.”
“Can’t you just check?”
“Why? If it mattered, it would tell me.”
More subtle still is behavioral correction. With 18 to 24 months of data, the AI knows which emails you ignore, which meetings you find valuable, which tasks you procrastinate on.
It starts predicting rather than responding.
“Based on your last seven meetings with Sarah, you felt they were unproductive. Should I decline future ones?”
“Yeah, that makes sense.”
The next month, it declines automatically and just notifies you afterward. You hesitate for a second, then think: it’s probably right.
The third year: reckoning
The third year is the moment of reckoning. An entire generation has outsourced memory. Not just phone numbers or directions, but everything.
“What did I promise Sarah?”
“What was I doing before this meeting?”
“What are my own goals again?”
People won’t know. They haven’t practiced remembering for years. Like a muscle left unused, memory atrophies.
Relationships begin to strain.
“You don’t even remember our life.”
“I do.”
“When is my birthday?”
The other person glances at their phone.
“June 14.”
“You just looked it up. You don’t actually know.”
The boundaries we probably won’t set
I think boundaries should be set before dependence forms. Medical decisions. Financial transactions. Anything involving children. Creative expression. These should never be handed to AI.
Once a week, we should review what decisions it is making for us, where our own agency has shrunk. Once a week, a fully offline day, just to preserve the ability to function independently. Not everything needs optimization. Some friction is healthy.
But honestly, we probably won’t do this. The benefits are immediate. The costs accumulate slowly. By the time we notice the problem, we’ll already be too dependent.
As I write this warning about AI’s darker implications, I’m using AI to help research it. That alone says everything. I’ve only been using it for 48 hours, and I already can’t imagine going back.
That should terrify me. It does terrify me.
Just not enough to uninstall it.
That is the real issue. Not that AI assistants are evil, but that they are too good. And anything dangerous that is too good to give up is, by definition, the most dangerous thing of all.
(This article represents the personal views and opinions of the author.)
