She's been doing legal research for eleven years at a mid-size firm that handles mostly civil litigation. She asked me not to use her name or the firm's name, which is fine — a lot of people ask that, and I think it says something about how much people are still figuring out what's okay to say out loud.
I asked her the question. Her sentence:
"It does the research in fifteen minutes that used to take me three hours, but now I spend most of my time proving to partners that I checked its work."
I asked her to say more about that — the checking part. She said the tools her firm uses are good at finding cases, citing statutes, building out background. Fast. Accurate enough that nobody's raised a serious objection to the output. But there's a new thing that's happened at the firm: partners don't fully trust the AI, and they also don't fully trust the paralegal to catch whatever the AI got wrong. So she's developed a kind of proof-of-work habit. She shows her notes. She marks the cases she cross-checked manually. She builds a paper trail that says: a human was here, a human looked at this.
She said she doesn't resent it exactly, but she does find it strange. She's doing more work to demonstrate that she's doing less work. The overall output is better and faster — genuinely. But the job is weirdly more anxious now than it was when she was doing everything by hand, because then at least the work was obviously hers.
I've been thinking about that ever since. There's a version of the AI-at-work story that's about replacement — the computer does your job, you go home. And there's a version that's about augmentation — the computer does the boring parts, you do the interesting parts. But what she described is a third thing: the computer does the work, and then you work to establish that you understand what the computer did, because without that proof, your role is unclear even to the people paying you.
That's not a good story or a bad story about AI. It's just what's actually happening in one office, to one person who's been at this for eleven years and is good at her job and is now spending part of every day annotating her own process for an audience of skeptical partners.
I asked if she thought the job would still exist in five years. She paused a long time, which I took as an answer, and then she said she thought it would look different. She didn't say more than that. I didn't push.