When AI Reveals Friction You've Normalized

Amy Westlake
Jun 2
3 min read

Situation

I had 387 items of feedback to synthesize.

Not 387 clean, categorized items. Raw entries - some overlapping, some contradictory, some that probably belonged to a completely different team. My job was to turn them into something actionable.

The way I'd always done this: read through every one. Spot the similar ones. Group them. Merge the descriptions. Flag the edge cases. Repeat until the (virtual) pile was manageable.

It was tedious. But it was also just... how you did it. There wasn't another option. So you got good at it, in the way you get good at anything you're forced to do repeatedly. You developed a system. You trusted your own pattern recognition. You moved on.

I'd been doing this for years. It never occurred to me to question whether I was doing it well.

The AI Move

I described the problem to Gemini. Not the feedback itself... the task. I have 387 items. Find the duplicates - not exact matches, similar ones. Merge the descriptions without losing critical details. Flag anything that's incomplete or that probably belongs to another team.

It did all of it.

The list went from 387 to 278. A hundred items collapsed, routed, or flagged. Descriptions merged cleanly. Nothing critical lost.

The time savings were significant. But that's not what stayed with me.

The Shift

What stayed with me was a question I couldn't stop turning over:

How had I been doing this effectively with my brain?

The honest answer: I hadn't been. Not really. I was doing it with my best effort, which is a different thing.

When you review 387 items manually, you're not processing them neutrally. You get tired. You start pattern-matching faster than you should. You push the ones you already agree with toward the top. You group things that seem related because of how they're worded, not because of what they mean. By item 300, your calibration has drifted from what it was at item 1.

That's not a personal failing. That's just how brains work under cognitive load. But I'd never had a comparison point before. There was nothing to put my process next to that would show me what it actually cost.

AI gave me that comparison point. And it was uncomfortable.

The Pattern

Friction you've normalized is invisible - until something shows you what it looks like from the outside.

The reason I never questioned my feedback synthesis process wasn't that it was good. It was that everyone around me was doing it the same way. Manual review was the only option, so manual review felt normal. The inefficiency was invisible because there was no contrast.

AI creates contrast. It doesn't get fatigued. It doesn't push its preferred interpretation of what "similar" means. It doesn't hit item 200 and start moving faster to get through the list. It just does what you asked.

When you see that output next to what you were producing manually, the question surfaces on its own: *wait, was I doing this right?*

Usually the answer is: you were doing it as well as a human could. Which is not the same as doing it well.

The Implication

The tasks worth questioning aren't the ones that feel broken. Those you already know about.

The ones worth questioning are the ones that feel fine - that feel like just how the thing is done. The ones where you've quietly developed a system, gotten decent at it, and stopped asking whether there's a better way.

Those are the ones where you've most likely adapted to the friction rather than solved it. And those are the ones where AI will hand you the comparison point you didn't know you needed.

Find one. Describe the task. See what comes back.

You might discover your process was perfectly calibrated. Or you might discover you've been relying on your brain for work it was never the right tool for.

Either answer is worth having.

What I'm Testing Next

The feedback synthesis was a single moment of recognition. But it changed what I look for now.

I'm less interested in tasks that feel hard - those get questioned naturally. I'm more interested in the ones that feel fine. Tedious but manageable. The ones I've built a quiet workflow around without ever asking if the workflow itself was the problem.

I'm starting to think "tedious but manageable" might just be what normalized friction sounds like.