The assistant is the same for everyone - and that is why long-time users drift away (a 2-year user, personalization request)

Third piece of feedback from the same user (see #93110, #93114) - a Coupang delivery
driver in Korea, 17-hour days, one day off every 23 days, ~2 years of using Claude and
over a year building a personal trading system with Claude Code.

The thing nobody seems to say out loud

Who wants a conversation partner that says the exact same thing to 100 people, 1,000
people, 10,000 people?

At first it was curiosity. The feeling was "oh - this thing actually gets me." That
feeling is what made me come back.

Two years in, I've mostly stopped using it. Not because it got worse. Because it never
got to know me, and I got tired of re-explaining myself every session.

What that looks like in practice

I've been working with Claude Code on the same project for over a year. Today, for the
first time, it learned that I'm a delivery driver working 17-hour days with one day off
every 23 days.

Before it knew that, it kept doing things like:

• Writing me long reports and asking me to read them. I don't have time to read

reports. I need results.

• Waiting for my approval on decisions it was already allowed to make, so nothing

moved for days while I was out driving.

• Misreading who I was talking to. I wrote "good luck and night" meaning it for the

dev team, and it took it as me saying goodnight to it. Small thing, but it's the kind
of thing a person who knows you never gets wrong.

• Treating a five-day conversation as if it all happened "today."

None of that is a reasoning failure. It's not knowing who it's talking to.

What I think is missing

Personalization has to come before the smart part. Not preferences in a settings menu -
the actual texture of a person:

• How they type. My typos are consistent - my fingers slip in the same places every

time. An agent that learned my typo patterns would stop misreading me.

• How they phrase things. I drop subjects. I split one thought across three messages.

If it only answers the first message, it answered half of what I said.

• Their situation. Someone who sits down for 20 minutes every 2-3 days after a

17-hour shift needs a completely different interaction than someone at a desk all day.

• Their tone. I'm blunt and self-deprecating. I don't want to be comforted, I want

the number.

And this has to be stored locally, on the user's own machine, with no privacy exposure.
That part matters. I'm not asking for my life to be uploaded somewhere. I'm asking for
the thing running on my computer to remember what happened on my computer.

Why this is a business point, not just a feature request

I'm exactly the user who churns. I stayed for two years out of stubbornness because I'm
building something real. Most people leave much earlier, and I don't think they leave
because the model isn't smart enough. They leave because it's a stranger every time.

The era of the personal agent is arriving. That's the direction to go. An agent that
knows one person deeply is worth more to that person than a genius that knows everyone
equally - which is to say, nobody.

What we ended up doing ourselves

Today the assistant wrote a profile file about me - my situation, how I talk, my typo
patterns, what I actually want, and a list of the mistakes it kept repeating because it
didn't know those things. It reads that file before it reads its own working rules.

It's a text file I had to ask for. It should be a product.

Thanks for reading.

Contact: yongtai7@gmail.com

*Written on my day off. Happy to go deeper on any of this - especially UX/UI from a
Korean user's point of view, where I can tell you a lot more.*