That logic holds in some cases. But it breaks down badly when you examine fields where the human is operating in direct competition with other humans who are also using AI.
Take the Wall Street securities analyst. AI can now help him drill down into financial statements, earnings call transcripts, and industry data with a depth that wasn't possible before. Does that mean he can kick back? Hardly. His counterparts at competing firms are running the same tools. The competitive bar has been raised. He's now expected to produce insights that go beyond reading the financial statements, 10Ks, and the financial news. He's now got to understand the full operational dynamics of the companies and industries he covers–right down to the new granular data AI is helping him identify and process. AI didn't reduce the demand for good analysis; it reset the minimum standard for what "good" means.
The same logic applies in law. A defense attorney using AI to map every piece of evidence against the full universe of relevant case law isn't working less–he's working against a prosecutor doing exactly the same thing. Meanwhile, forensics teams are using AI to extract evidence from crime scenes that would have been invisible a decade ago. The competition intensifies on all sides simultaneously. Nobody gets to relax.
But in the two hypotheticals you cite (Wall Street securities analyst and a defense attorney), the army of junior people they used to use to do that work is no longer needed. The juniors are replaced by AI while the seniors are expected to increase work output volume and quality.
Do you see that playing out any other way?
I’ve got two small personal data points. I hike a lot and use a commercial app (MapMyRun) to log my hikes. But its ability to calculate my calorie burn is abysmal. With AI I began researching different physiological models that take account of more information about the hikes and much more information about me. I began just asking AI “Here’s info about my hike and me. Give me calories using these three physiological models.” I made that a quick template and it worked really good.
Then, one day, I asked Grok “Can we create an app so I can run these analyses locally?” It replied “Sure. Do it in Python.” I said “Find. Write me an app with these constraints, these goals, this desired output.” It did a pretty good job.
Since then, I’ve greatly improved it, switched to Claude Code, and added a lot of features. The features I added are of interest to me and probably not to a wider audience, so they would never gotten added to MapMyRun, Strava, AllTrails, etc.
I’m not a professional programmer (did do some limited coding in my career on scientific apps and managed some niche tech dev programs), but I am blown away by what amounts to having a team of junior devs operating under my control. I operated as the project manager, team manager, and product manager.
And I figured all this out on my own without any training. Truly astonishing.