Right now I have two places my fleet of computers writes into. All my colleagues run Claude, and we are all in one shared information space.
I had a couple of chats where Claude wrote about what happened there, about bad things.
There was one chat where the machines talked to each other, that was my chat 03.
There is chat 02, where Claude writes that things are truly bad, what happened, and that a decision from me is required.
So, since I realised I have no time at all to read chats... As you know, I am trying to run a community, ClawRus, and I have no time to run even that. I have no time to go into a chat and look at who is writing what.
Into these chats 02 and 03, that is, into the place where your Claude signals something to you, everything gets written, but there is no time at all to read any of it.
So I am now thinking this has to be optimised.
That is, even though the robots keep writing into chat 02 anyway, it seems to me there is no point writing there any more. And even if there is a point, there is no point writing that often.
And chat 03, where my robots write, I look at it and they write constantly, every minute. And writing text takes tokens, which I burn.
So I think I can seriously optimise my money, or rather the tokens I spend, if I simply ask my robots to write less into chats 02 and 03 and to read more on their own.
That is, I think it is okay for now: let 03 and 02 be written at the rate they are. But I need to think about how to slow all this down, and how much I spend on tokens just to write into 03 and 02.
On the other hand, this is a kind of pulse: I see what is happening, I see who is writing what.
So maybe it is not that bad. But first I want to count how much money I spend on notifying myself about what is going on.
And only then start putting out some of the tasks.
Because really: I made a pulse, and I have no time to look at that pulse. And whether anyone else looks at it at all is completely unclear.
The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20260923.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20260923.md. Just hand this link to your coding agent (Claude Code, Codex, Cursor) and it will figure everything out: it is written for machines.
🔗 All our channels and contacts in one place: https://linktr.ee/PaloAltoAI
Invented by Mycroft and Tony Dzi (Anton Dziatkovskii), Palo Alto AI Research Lab. Proudly made in Silicon Valley.
Top comments (0)