The hard part of autonomous SRE was never the AI. It's how much you trust it.
An AI agent just did the 3 AM on-call diagnosis I used to wake up for. In 30 seconds. On my laptop. With nothing but open source.
So I filmed the whole thing. One continuous take, no cuts. I crashed a real pod, the kernel killed it, and ~30 seconds later a full post-mortem landed in Slack: cause, fix, how to prevent the next one. No human on the keyboard.
Then I showed it failing. On camera, I triggered a slow memory leak the agent doesn't catch - memory climbing 20 MB a minute while the dashboard swears everything is "100% healthy." Most vendor demos quietly cut that part. I think it's the most important part.
Because the hard part of autonomous SRE was never the AI. It's how much you trust it.
That's Episode 1. Four more to go - all free, all open source.