Could the race to build human-level AI be crowding out AI that invents entirely new kinds of work?

This explores whether chasing AGI steers firms away from AI that creates new tasks for people, as opposed to AI that replaces the tasks people already do. The corpus doesn't measure where firms actually spend their R&D money, so it can't settle the economic question. It does give a sharp account of how an AGI goal shapes what researchers see as worth building, and some field evidence on what happens when AI takes over existing tasks.

The most direct material is a critique of AGI as a guiding goal for research Does treating AGI as a north star goal undermine research planning?. It names six traps. Two of them bear on this question. A "goal lottery" means the aims that win attention are the ones that fit the AGI story, not necessarily the ones that would help most. "Generality debt" means specific, concrete problems get put off because they seem too narrow next to the big prize. Building AI that opens new kinds of work for people is exactly that kind of specific, domain-by-domain project. The paper's fix is to name concrete goals, accept several competing goals at once, and include more people in choosing them. The AGI-to-superintelligence roadmap shows the same habit from the opposite side What bottlenecks define the path from AGI to superintelligence?. It maps the future entirely as routes to more capable systems. In that picture, human work appears as friction, not as something to design for.

Other notes suggest that even without the AGI framing, the default is to swap AI in for human work rather than expand it. When AI does the engaging parts of a job, workers are left with the dull leftovers and say they feel less motivated and more bored Does AI collaboration drain motivation when workers return to solo tasks?. That is automation with no new tasks to replace what was lost. A related argument is that AI now produces the finished form of intellectual work (the essay, the design, the plan), separated from the reasoning that used to come with it Does AI separate intellectual form from the thinking behind it?. If firms put money on that capability, they are betting on substitution.

The counterexample is useful. In a field experiment at Procter & Gamble, individuals using AI matched the performance of two-person teams. They also produced solutions that drew more evenly on different professional backgrounds Can generative AI replace the benefits of having a human teammate?. That points toward AI that changes what one person can take on, which comes closer to creating new tasks. Benedict Evans explains why this kind of payoff is hard to get: making tools easier to build doesn't help if workers can't see which of their tasks AI could change, or if companies can't coordinate the change across departments Does easier tool-building actually solve enterprise adoption problems?. Inventing new work is an organizational design problem, and a race toward general capability doesn't address it.

What you might not have expected: the corpus suggests the AGI goal may do less harm by pulling money away than by narrowing imagination. It makes the most general system look like the only goal worth having. Meanwhile the slower work of figuring out what new things people could do with AI gets treated as a side issue for deployment, not as research. For the economics itself, meaning firm incentives and the balance between automating tasks and creating them, this collection is thin. You'd need labor-economics sources beyond it.