AI generates research ideas faster than I can test them. Anyone else? [R]

I've been using AI extensively for ML research, and I've noticed a bottleneck: brainstorming research directions and generating smoke run implementations has become incredibly fast, but actually testing those ideas hasn't. A single AI discussion (eg asking it to review 30 papers I had in my reading list, audit them, check for ideas related to the status of my project) branched into dozens of promising approaches, and with coding agents, exploring them in parallel is in principle feasible. But as a researcher, I still have to manage experiments, launch training runs, evaluate results, and decide what to try next. And, importantly, to understand what's gong on. The result is an overwhelming number of potentially interesting ideas that I simply can't keep up with, risking to not really knowing what should be the smartest "next step". I'm wondering if others are experiencing the same thing. Have you found good ways to manage this, or best daily practices that help in this era of "vibe-research"?