Aug 10, 2026 08:13 PM
https://www.sagepub.com/explore-our-cont...es-experts
EXCERPTS: A first-year associate reads through discovery documents. A junior developer fixes small bugs and writes boilerplate. A new analyst builds the routine slide deck nobody senior wants to build. None of that work is glamorous. Most of it is now automatable, and organizations are automating it. From an operational standpoint, they are right to. The work is repetitive, the AI is fast, and the savings show up this quarter.
But that work was never only production. It was also how people learned the job. The associate reading discovery is building a feel for what matters in a case. The developer fixing small bugs is learning how the system actually behaves, as opposed to how the documentation claims it behaves. You cannot get that from a training course. You get it by doing the thing badly, many times, until you stop doing it badly.
Remove the work, and you keep the output. You lose the learning that came with it.
[...] There is a second problem that concerns me more than the first.
Nearly every serious proposal for governing AI depends on human oversight, on somebody in the loop who can tell when the machine is wrong. That only works if the person has independent expertise. Someone who has always worked with AI assistance, and never built the underlying judgment, cannot really audit the system they rely on. They can only agree with it, confidently and quickly.
So oversight is not a resource we can assume will be there. It is produced. And it is produced by exactly the work being automated. That is the connection I think the AI governance conversation is missing.
[....] So what do you do about it? If you treat this as a training problem, you get training solutions: more courses, more upskilling, more internal academies. Those are worth doing, and they will not be enough, because no organization can solve a collective action problem by acting alone.
[...] Expertise is not collapsing. But current incentives are pointed in a specific direction, nobody is watching where they lead, and the people best positioned to notice are the ones we are on track to stop producing... (MORE - missing details)
EXCERPTS: A first-year associate reads through discovery documents. A junior developer fixes small bugs and writes boilerplate. A new analyst builds the routine slide deck nobody senior wants to build. None of that work is glamorous. Most of it is now automatable, and organizations are automating it. From an operational standpoint, they are right to. The work is repetitive, the AI is fast, and the savings show up this quarter.
But that work was never only production. It was also how people learned the job. The associate reading discovery is building a feel for what matters in a case. The developer fixing small bugs is learning how the system actually behaves, as opposed to how the documentation claims it behaves. You cannot get that from a training course. You get it by doing the thing badly, many times, until you stop doing it badly.
Remove the work, and you keep the output. You lose the learning that came with it.
[...] There is a second problem that concerns me more than the first.
Nearly every serious proposal for governing AI depends on human oversight, on somebody in the loop who can tell when the machine is wrong. That only works if the person has independent expertise. Someone who has always worked with AI assistance, and never built the underlying judgment, cannot really audit the system they rely on. They can only agree with it, confidently and quickly.
So oversight is not a resource we can assume will be there. It is produced. And it is produced by exactly the work being automated. That is the connection I think the AI governance conversation is missing.
[....] So what do you do about it? If you treat this as a training problem, you get training solutions: more courses, more upskilling, more internal academies. Those are worth doing, and they will not be enough, because no organization can solve a collective action problem by acting alone.
[...] Expertise is not collapsing. But current incentives are pointed in a specific direction, nobody is watching where they lead, and the people best positioned to notice are the ones we are on track to stop producing... (MORE - missing details)