The comfortable version of this argument goes like this: AI compresses a task inside a job. It doesn't eliminate the job. So the same people get freed up for higher-judgment work. I've built the systems that argument describes: research agents, intake automation, matching engines that used to be manual screening. I don't think the comfortable version holds up once you follow it past the first team it hits.
Task compression is real. I've seen it work. What I don't buy is the next step: that a company facing a compressed task keeps the same people and gives them more interesting work. That's not the default response when a cost center gets cheaper to run. The default response is to run it with fewer people.
Why the Freed Capacity Doesn't Automatically Turn Into More Output
The task-compression argument assumes demand for a team's output grows fast enough to absorb the capacity AI frees up. Sometimes it does. A lot of the time it doesn't. Demand for first-pass funding research or initial candidate screening isn't that elastic. There's a roughly fixed amount of that work a business needs done. Once an agent can do it faster, the business doesn't need more of it. It needs the same amount with less labor cost. That's not a failure of the technology. It's the technology working as intended. And the intention was never really "give junior staff more interesting work."
Entry-Level Roles Are the Most Exposed, Not the Least
This is where the optimistic framing gets it backwards. A lot of entry-level roles exist to do the legwork these systems now compress: first-pass research, initial screening, routine documentation. That isn't a small inconvenient part of the role sitting in front of the "real" job. For a junior person, it often is most of the job. It's also how they build the judgment that eventually lets them do more senior work.
Compress that task hard enough and you haven't freed up a junior person's time for better work. You've removed most of the reason the role existed. That's a very different outcome than the task-compression story describes. It's also the one I expect to show up first: fewer entry-level seats. Not the same number of juniors doing more strategic work.
Support Functions Are the Same Story With a Different Name
The same pattern applies to support and back-office roles. A lot of that work is mostly the routine task, not a routine task sitting beside a bigger judgment-heavy job. When the routine task is most of the job, compressing it doesn't leave much of a role standing. I think that's a bigger share of the labor market than the optimistic framing admits. Knowledge-work commentary tends to focus on roles that still have judgment left over after automation. It quietly ignores the much larger set of roles that don't.
Why I Think This Gets Undersold
Part of it is caution: nobody wants to predict mass layoffs that don't show up on schedule, so the safer public position is the moderate one. Part of it is incentive. Companies building and selling these systems prefer "augmentation" over "substitution," because "this replaces headcount" is a harder sell than "this makes your team more effective." That's not a conspiracy. It's a predictable framing bias. And it has shaped the public conversation more than the underlying economics actually support.
None of this means the technology shouldn't be built, or that task compression isn't useful. It means I'd stop assuming the labor-market cost shows up gradually and gently. I think it shows up concentrated in the roles that were closest to pure routine task to begin with. That's a harder story than the one currently being told.