Can AIs self-improve? How? Then what?

I've been thinking a lot about the idea of recursive self-improvement, where AIs are used to improve AIs. Is it possible? What shape does it take? Is it something we can do or make?

Have some ideas I'm burning to share and scrutinize. Hoping for other pairs of ears and eyes! Of course, your brains and speaky-parts will be welcome, too!
So bring ideas while we enjoy this Springtime weather outside drinking and/or eating Central Market North Lamar Cafe's wonderlicious food!

Also open to talking about other related or non-related nerdy topics. Lately, in my world:
- Created an agentic coding Team. Using it to build apps.
- Installed and using an OpenClaw.

A conversation I had yesterday with two AIs about this topic:
https://clearsay.net/framework-for-ai-self-improvement-via-flaw-injection/