Does cognitive load explain anything, or just describe what we already knew?

I keep running into this in product conversations, usually framed as a solved problem. We need to "reduce cognitive load" on users. Sounds right. But when I actually try to trace back what's being predicted—not described, predicted—it gets slippery fast. The concept comes from Sweller's work in the 80s on instructional design, which was genuinely useful: the observation that working memory has limits, and presenting information in ways that respect those limits improves learning. Fine. But somewhere between that specific claim and "make things simpler because brains," the framework became decorative. People use "cognitive load" the way they used to use "intuitive" — as a vague nod toward something that feels true. You don't have to commit to a mechanism. You just say the interface has high cognitive load and nod knowingly. Here's what troubles me. The framework doesn't cleanly predict which simplifications actually matter. You can remove options, reduce text, flatten hierarchies, and—depending on context and user intent—performance either improves, stays flat, or tanks. I watched a team strip down a workflow to fight "cognitive overload" and destroy its utility because power users suddenly couldn't do their jobs. Was that because working memory wasn't the constraint? Because the constraint was *information scarcity*, not excess? The framework doesn't have a clear way to diagnose that before you ship. The medieval parallel keeps coming back to me. Before germ theory, you had a decent observational rule: don't leave wounds open, don't let the sick breathe bad air. Sound practice built on half-right intuition. Then microscopes arrived and you could actually see what was happening. We're still mostly in the intuition phase with cognition. Cognitive load might be real. But the framework predicts almost nothing you couldn't predict by just watching people and asking what's hard about their task.