Self-improvement of a session.
What is self-improvement of a session?
Say you have a routine that submits your candidacy to 10 job openings every day. That routine opens various boards to apply you to positions, and those boards, I do not know, maybe something falls off on them and so on. Some boards, the very format of those boards changes, the places where vacancies are posted.
So, the routine improves itself.
Every routine, it seems to me, is obliged to improve itself.
That is, every time it works out what it tripped over, tries to solve that, and rewrites itself into the future, makes itself better. I think it is a great thing.
Not necessarily only when applying you to jobs, there are many use cases.
Any routine, I do not know, say you have a content factory, the content factory handles posting and distribution, and it improves itself so that in future it can be more optimal, more productive and make fewer mistakes.
I do not know how well this works on models like Sonnet or old simple ones like Luma and so on, but on Opus and on Fable such routines work just perfectly, they improve themselves. It will not get you to AGI, of course, but you really do spend less time keeping the infrastructure alive.
So a routine is obliged to improve itself.
That is a genuinely mandatory criterion.
If you have already learned to decompose all of your daily tasks into a pile and a pile of routines, and you make them self-improving, you will save yourself a very large amount of time.
If anyone needs self-improving routines, I have, I do not know, getting on for a hundred routines, if not two hundred, I do not know how many, maybe three hundred, I have not counted, but I really do have a lot of them, then I am ready to share with you.
If your Claude tells you what your pains are, we can make a little chat right now where my Claude talks to your Claude, and we will try to fix it. More precisely, not fix it, I will simply hand over all the knowledge I have so that you can use Claude, Codex and whatever more optimally, to be more productive.
Once again, any skills and so on have to be self-improving, because something is constantly falling over, falling off, there is the API, MCP, sites change and so on, you constantly need to improve yourself.
This will not bring you to AGI, but it will bring you to your routine surviving. You do not have to keep running around wiping noses.
Imagine, I have 300 routines there, I would not be able to run around wiping all their noses if I did not have this self-improvement, this self-improvement skill.
The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20261022.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20261022.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)