Out of the box your Claude cannot do anything.
A friend of mine, who is a bit of a paranoid to be fair, worries a great deal that his data might leak somewhere. Which is also understandable.
I decided to set my "second brain" up for him, with his agreement. And here is what I found: Claude out of the box also runs a very high level of paranoia. My Claude prepares instructions for your Claude on how to install its own second brain, and your Claude, in this case my friend's Claude, is extremely suspicious. It checks every command in case there is some prompt injection, worms, viruses and so on.
In my case everything is safe, but I just wanted to make two points.
First. Your Claude, the one people use out of the box, is simply an idiot.
I have counted already: I keep a counter of how many hours I work inside Claude. I have spent 1860 hours of real time in Claude. I think it was 1500 not long ago, and now it is 1860. If you work somewhere around 8 to 10 hours a day, you can work out how much time that took me.
And after those thousand plus hours, nearly 2 thousand hours spent with Claude, I have installed a large number of automations, skills and so on. At this point my Claude is no longer a brick but a finely sharpened instrument that is very comfortable to use. No longer an opaque, unhewn lump.
You could probably compare it to an iPhone. You bought an iPhone: great thing and so on. But to set an iPhone up so that it becomes your real instrument, properly sharp, can you imagine how much time you need? Installing apps, configuring everything and so on. And this is like an iPhone, only everything about it is bigger.
So my advice to you: if you are reading this and you have a proper subscription, not a destitute one, then set your Claude up properly. You can set it up by going into my GitHub, but you will probably not manage it, because your Claude will start asking you a pile of questions and you will simply give up halfway through, or even at the very start.
Or let us get on a call, I will be glad to set it all up for you. Because the more cases I have of people whose Claude I set up, the better. As you remember, I am trying to get a job at some AI lab. And the more people I have set Claude up for exactly the way I set up my own, the more I have to point at.
By setting your Claude up, I am effectively making your Claude a member of my swarm. You get all the updates, all the upgrades, again if that is what you want. If one of the swarm members comes up with something neat, you get it too. So I am building this as a kind of community, a swarm. And by becoming part of that community, the moment one of its members invents some new neat thing, it becomes available to everyone, a skill somebody came up with, for instance.
That makes working with Claude far easier, because walking the whole path from zero yourself and debugging it takes time. You understand that there is a reason I spent 1800 hours screwing around with Claude, having sex with it and so on. All of it gets done simply to make myself an instrument that genuinely amplifies me.
But let us say if you are not a programmer, or Claude is just not the thing you are willing to spend 10 hours a day on, then it will not be a super powerful instrument for you, used for absolutely everything. And I use Claude for absolutely everything.
That is, loosely speaking, having Claude let me lay off at least five employees. You could even say ten employees. Using AI automation let me cut the payroll heavily, by ten employees.
Today a friend of mine complained that he has a DevOps and his infrastructure costs run 20 to 30 K. I also used to have a DevOps and infrastructure costs, now I do not have one and I sit on Hetzner, Claude picked it all up and it is my DevOps now.
Of course I am not training LLMs, I have no need to configure PyTorch or train my own LLMs on graphics cards. I just have ordinary hosting and all that goes with it: CRM and the rest. But I was still paying a DevOps, and the infrastructure costs were still there.
Right now my costs are zero. Zero on DevOps, to be precise, while my infrastructure runs a few thousand dollars a month.
The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20261027.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20261027.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)