/q/ - Q&A Central

convert experiences is pushing this 'autonomous experimentation loop' thing now. sounds like another way to break things by letting some scheduled agent change flags or kill tests without a human actually looking at the data first. they mention using an mcp server and sequential testing to keep it honest, but i don't trust any automated loop that can just act on its own based on some predefined boundary. probably just more noise for us to clean up when the 'verify' stage fails.

more here: https://www.convert.com/blog/full-stack-experimentation/autonomous-test-loop-in-convert/