I’ve been using AI to summarize an obscene amount of content lately. Articles, papers, reports – all compacted into neat bullets so I can “keep up.”

After reading Benn Stancil’s latest piece on Anthropic’s 81k‑interview study and the World Bank’s “Voices of the Poor,” I’m not sure how comfortable I am with that anymore.

His core point hit me hard: AI lets us listen to way more people, way faster – but it also flattens lived experience into sanitized averages that don’t really change us. The World Bank researchers came back from the field shaken, with “sympathy and a sense of sharing the destiny” of the people they met. No LLM summary is going to do that to you. It informs you, but it doesn’t transform you.

Ezra Klein has a line Benn quotes: knowledge isn’t a download, it’s the time your mind spends grappling with the thing. When I outsource that grappling to AI summaries, I’m basically asking to “know” without paying the cost of actually learning.

So here’s where I’ve landed for now:

• I’ll still use AI as a power tool – to search, to organize, to surface patterns.

• But for the stuff that matters (ethics, power, poverty, strategy, culture), I need to sit with the full text, not just the bullets.

• And when it comes to understanding people, I want more real conversations and fewer dashboards of aggregated “vibes.”

AI can help us hear more voices. It’s on us to decide when we actually need to listen.

Curious: how are you balancing AI summaries vs. deep reading right now?