Maybe I'm wrong, but at least on the internet, it feels like we're trying to minimize cognitive friction at all costs.

I suspect that's because most major platforms incentivize attention and engagement at the expense of trust and quality.

I think that's very misguided.

Cognitive friction, the kind that comes from actually doing the work of thinking, is one of the most useful signals I know that I'm doing something worthwhile. Yes, it doesn't usually feel great in the moment, but it's a pretty good proxy for growth and learning.

Not all friction is like that. Friction from tedium, where you're doing something rote or mechanical, isn't very helpful. But friction from cognitive labor is worth paying attention to.

The labor is the thing that gets distributed in a social interaction. The friction is what you feel when you're doing your share. Skip the labor and you skip the signal, but the labor doesn't disappear. It just gets pushed downstream.

This skewed distribution can quickly erode trust.

An example.

Imagine someone is trying to submit a patch to an open source project.

In world one, they use their AI agent to do the work, it generates slop they don't personally review, and they submit the patch.

In world two, they do the work themselves and submit the patch.

In world one, the contributor felt no friction because they didn't do any labor. All of it lands on the maintainer, and it's not the good kind. It's pure tedium. Assuming they give a shit about quality, as soon as they realize it's slop, they reject the patch and the interaction is over. Trust with the contributor is gone. No one learned anything.

In world two, the cognitive friction is more evenly distributed. The person had to put time and energy into understanding the codebase. They felt the friction. Even if they made a mistake, their initial investment will make the maintainer more likely to give feedback. The contributor learns something. The maintainer gains another person who understands their project. Trust is built.

This isn't an anti-AI post. I think there's a version of world one — where the contributor takes some time to understand the codebase and review their agent's work — that gets pretty close to world two. You can still feel cognitive friction and use AI. I just think it's very tempting not to.

That's a slippery slope. You're cheating yourself out of an opportunity to learn, and you run the risk of eroding trust.