Emily, this is really interesting to me, particularly your observation that there seems to be a need for generalist skills that isn't necessarily visible in the job market.
I wonder whether a source of operational freelancers could be a useful resource for EA—and potentially AI safety organizations specifically. A pool of experienced people who could come in for defined projects, fractional work, or short-term operational capacity could lower the barrier on both sides.
But I'm also wondering whether the bottleneck is actually less about a lack of talent and more about talent recognition and organizational design.
A lot of what you're describing sounds very familiar from the startup and growth environments I've worked in. Organizations know they have too much to do, but they don't always know exactly what role they need, how to structure it, or what capabilities should sit together. That's an organizational design problem as much as a recruiting problem.
I'm particularly curious about the emphasis on “context.” Obviously, there is domain-specific context that matters enormously for technical research and policy work. But for operations, I'm less convinced the amount of context required is as high as it sometimes seems to be.
The basic building blocks of building and scaling an organization—clarifying priorities, designing roles and decision rights, establishing processes, creating operating rhythms, improving systems, identifying bottlenecks, managing projects, building capacity—have a remarkable amount of commonality across industries.
So I keep wondering: how much context is actually necessary before an experienced operator can start producing useful work, when that work is neither technical nor research in nature?
I've actually started considering offering myself as a freelancer into the ecosystem for exactly this reason—not necessarily to arrive knowing what the organization needs, but to help an organization figure that out. Come in, understand the work and the bottlenecks, help define what capacity is missing, determine what the roles should look like, and then either help build that capacity or hand off a much clearer hiring plan.
It seems like that could be a useful bridge between “we need more capacity” and “we have a perfectly defined job description for you.”
And your point about the amount of unpaid work required to establish credibility makes me wonder whether we're inadvertently filtering out some of the very experienced people the field needs. Not everyone who has valuable operational expertise is going to be able—or willing—to spend six months proving that they belong before they're allowed to use it.
Anyway, I'm still thinking this through, but I really appreciated you articulating the problem so clearly. It has me looking at the generalist bottleneck less as a talent-supply problem and more as a possible talent-routing + organizational-design problem.
This was a very thoughtful response, Chris. My only feedback would be that I’m curious about your idea, or anyone’s idea, of “cultural fit.” There is a LOT of room for personal bias in that assessment. And the more I think about it, the more I wonder whether “cultural fit” can sometimes become a proxy for “this person thinks and communicates like the people already here.”
Given the importance of what is being built, and the reasons why, perhaps hiring managers should periodically re-examine which characteristics are actually predictive of impact—and which are simply familiar or comfortable.
Especially if the field is struggling to absorb experienced people, that seems like a question worth asking.
Emily, this is really interesting to me, particularly your observation that there seems to be a need for generalist skills that isn't necessarily visible in the job market.
I wonder whether a source of operational freelancers could be a useful resource for EA—and potentially AI safety organizations specifically. A pool of experienced people who could come in for defined projects, fractional work, or short-term operational capacity could lower the barrier on both sides.
But I'm also wondering whether the bottleneck is actually less about a lack of talent and more about talent recognition and organizational design.
A lot of what you're describing sounds very familiar from the startup and growth environments I've worked in. Organizations know they have too much to do, but they don't always know exactly what role they need, how to structure it, or what capabilities should sit together. That's an organizational design problem as much as a recruiting problem.
I'm particularly curious about the emphasis on “context.” Obviously, there is domain-specific context that matters enormously for technical research and policy work. But for operations, I'm less convinced the amount of context required is as high as it sometimes seems to be.
The basic building blocks of building and scaling an organization—clarifying priorities, designing roles and decision rights, establishing processes, creating operating rhythms, improving systems, identifying bottlenecks, managing projects, building capacity—have a remarkable amount of commonality across industries.
So I keep wondering: how much context is actually necessary before an experienced operator can start producing useful work, when that work is neither technical nor research in nature?
I've actually started considering offering myself as a freelancer into the ecosystem for exactly this reason—not necessarily to arrive knowing what the organization needs, but to help an organization figure that out. Come in, understand the work and the bottlenecks, help define what capacity is missing, determine what the roles should look like, and then either help build that capacity or hand off a much clearer hiring plan.
It seems like that could be a useful bridge between “we need more capacity” and “we have a perfectly defined job description for you.”
And your point about the amount of unpaid work required to establish credibility makes me wonder whether we're inadvertently filtering out some of the very experienced people the field needs. Not everyone who has valuable operational expertise is going to be able—or willing—to spend six months proving that they belong before they're allowed to use it.
Anyway, I'm still thinking this through, but I really appreciated you articulating the problem so clearly. It has me looking at the generalist bottleneck less as a talent-supply problem and more as a possible talent-routing + organizational-design problem.
This was a very thoughtful response, Chris. My only feedback would be that I’m curious about your idea, or anyone’s idea, of “cultural fit.” There is a LOT of room for personal bias in that assessment. And the more I think about it, the more I wonder whether “cultural fit” can sometimes become a proxy for “this person thinks and communicates like the people already here.”
Given the importance of what is being built, and the reasons why, perhaps hiring managers should periodically re-examine which characteristics are actually predictive of impact—and which are simply familiar or comfortable.
Especially if the field is struggling to absorb experienced people, that seems like a question worth asking.