Recently I've been asking friends what their post-AGI job would be. I get mixed responses - everything from "Oh I'll just keep working in research at company X" all the way to "Hunting".

Pre-AGI jobs

Before we get there, there's an interesting window we're about to enter (or are already in). AI surpasses humans at cognitive work but robotics hasn't caught up. Humans become the physical layer.

Sholto Douglas and Trenton Bricken from Anthropic on the Dwarkesh Podcast 1: humans with AirPods and smart glasses, directed by AI through cameras, told what to do step by step. "Basically, you're having human meat robots". This is already happening. RentAHuman.ai 2 is a marketplace where AI agents directly hire humans for physical-world tasks: attending meetings, picking up packages, field research. The pre-AGI jobs of the near future will most likely be well-paying too. Transporting materials, in-person meetings on behalf of AI systems, travelling, contracting physical work.

How long does this window last? Todd estimates car factories could be converted to produce a billion robots per year in under five years 3. Aschenbrenner argues that once AI automates AI research, solving robotics follows quickly 4. Either way, for some period the most valuable human skill is having a body and being somewhere.

Post-AGI jobs

Contrary to what even some BigLab CEOs will say publicly 5, AI will replace most of the workforce. Both model intelligence and robotics are rapidly converging, and might even need each other to truly succeed. I think displacement affects three types of people differently:

  1. Slaves to the system who hate their job. These people will feel liberated and will finally have the time to do what they want, spend time with their families, pursue hobbies and enjoy life.

  2. Slaves to the system who love their job. They know how to play the game, rose through the ranks, earned their title. In this process they learned to love the system, they are actively a part of it. But the ranks are meaningless now. Most will not cope well with this.

  3. Not quite slaves to the system. Institution-adjacent but free to pursue ideas, mostly research and academia. Their identity comes from the work itself. Similar situation as 2. but maybe more existential. Their work does not directly feed into the system to make it perpetual, but relies on "a" system, maybe the system of ideas. You spent your career on a problem and something solved it overnight.

The moral dilemma: are you researching your topic because you love it, or because it creates a net positive for society?

Figure out how to produce more than you consume. If you do that you're a good person. If you don't do that then you're a bad person.

  • George Hotz

Not to be taken this harsh, more as a heuristic. If you say you do research for humanity, you should welcome being replaced by something faster and better. If that thought horrifies you, it was never about humanity.

Geoffrey Hinton's career advice? Be a plumber 6. Cheng wrote a taxonomy on LessWrong of jobs resistant to AI automation 7: construction workers, surgeons, firefighters, custom furniture makers. The common thread: specialized robotics still cost more than a person. Moravec's paradox holds 8.

So, what's mine?

I had long thought about this question and it comes down to two things for me: ski instructor or hiking trail pathfinder. Maybe even dog walker - or robot dog walker - whatever.

Footnotes

  1. Dwarkesh Podcast - Sholto Douglas & Trenton Bricken. ↩

  2. RentAHuman.ai - "The meatspace layer for AI." ↩

  3. How Quickly Could Robots Scale Up? - Benjamin Todd. ↩

  4. Situational Awareness - Leopold Aschenbrenner. ↩

  5. See for instance Dario Amodei's and Demis Hassabis' public statements on AI and employment. ↩

  6. Geoffrey Hinton, on being asked what jobs will remain safe from AI. ↩

  7. A Taxonomy of Jobs Deeply Resistant to TAI Automation - Deric Cheng. ↩

  8. Physical manipulation requires orders of magnitude more compute than cognitive tasks. See Moravec's Paradox and AGI Labor Markets. ↩