A friend told me he'd shared a link with his buddies at work. "Every morning we have to start our chats from zero."
That sentence has stayed with me.
Not because it's new. Anyone who's used Claude or ChatGPT as more than a search box has felt it. You build context all morning — who you are, what you're working on, who the people in your life are. By the time the AI understands enough to be useful, the conversation is rich. You close the tab. The next day you open it. The room is empty.
He said it casually. Like the weather. Just something everyone deals with.
The phrase is in the wild
I went looking. The exact phrase — start from zero every morning — is being said in public, by people who don't know each other, in roughly the same register.
A Twitter post from yesterday, 170+ likes: "Every single time. Claude forgets them overnight. They start from zero every morning. They retype the same context. They re-explain their voice."
A Substack writer: "Every day is their first day."
A Medium piece: "A brilliant assistant who gets amnesia every morning."
A YouTube commenter: "Cut off mid-workflow. One message and I'm done."
Same metaphor. Same weariness. Same workplace experience, scaled to whoever is at the keyboard.
When language converges like this — strangers independently arriving at the same words for the same feeling — it usually means the feeling is real and the words are right.
What's actually happening
The technical name is compaction. The chat history gets long, the model can't carry it all, so the system summarizes the older parts to make room. Sometimes the summary is good. Sometimes it isn't. Eventually, even the summary gets summarized. Past a certain point, the original context is just gone.
The AI doesn't lose you because it doesn't care. It loses you because there's nowhere for the caring to live between conversations.
This is a structural thing, not a Claude thing or a ChatGPT thing. Any chatbot built the same way will have the same problem. Better models will hit it slower. None of them solve it by being smarter — they solve it by being bigger, until you hit the new limit and the cycle repeats.
The problem isn't intelligence. The problem is where memory lives.
What the audience already knows
Here's what struck me reading those quotes: nobody is asking for a smarter AI. They're asking for one that knows them by now.
That's a different request. Smarter AI is what vendors are selling. Knows me by now is what users are missing.
The fix isn't to make the same chat smarter. The fix is to put memory somewhere that doesn't get reset when the chat ends. A place outside the conversation that the conversation can read from and write to.
Once memory lives there, every morning isn't from zero. The AI walks in already knowing what mattered yesterday. The ten minutes of re-explaining disappear. The relationship compounds.
This isn't a feature. It's a different shape for the tool.
The quiet part
I think the reason that sentence stayed with me — every morning we have to start our chats from zero — isn't the pain it names. It's that he said it without complaint. Like he'd accepted it.
That acceptance is the most expensive part. Not the lost time. Not the redo. The slow giving up on the idea that the AI could ever actually know you.
A tool that resets nightly trains its user to expect nothing of it that requires continuity. So we stop asking for continuity. We adapt downward. We do the work the tool should do.
We shouldn't have to.
methinks the conversation isn't the relationship. The substrate is.