The unit of productive intelligence is shifting from the individual to the human+AI configuration, so the useful question stops being how capable the AI is and becomes what the coupling is like. A Stradivarius in clumsy hands produces less than a student violin played with full attention. What that leaves standing is Arendt's older question: labour is what gets automated first, and work and action are what the door opens onto, if you can reach the door at all. The chapter answers in play rather than in theory, with a game built at two in the morning that none of the three who made it could have made alone.
-
1
Hannah Arendt, The Human Condition (1958)
-
2
Irvin Yalom, Existential Psychotherapy (1980)
-
3
Erich Fromm, The Sane Society (1955) and Escape from Freedom (1941)
-
4
Marc Andreessen on the job as a bundle of tasks
-
5
Erik Brynjolfsson et al., Generative AI at Work (QJE, 2025)
-
6
Robert Allen, Engels’ Pause (Oxford Economic Papers, 2009)
-
7
Andrej Karpathy on orchestrating agents, and on the app he built properly and then called spurious
-
8
The Stradivarius Principle and the Tennis Paradox
-
9
Naval Ravikant on the decisions we make by default
-
10
Nicky Case, The Evolution of Trust (interactive)
-
11
Robert Axelrod, The Evolution of Cooperation (1984)
-
12
Silver et al., Mastering the game of Go with deep neural networks and tree search (Nature, 2016)
-
13
Demis Hassabis on the Lex Fridman Podcast #475
-
14
Rosa Fioravante and Antonino Vaccaro (eds.), Humanism and Artificial Intelligence (Springer, 2025)
-
15
Pedreschi et al., Human-AI Coevolution (Artificial Intelligence, 2025; preprint 2023)
-
16
Pilot Quest, the game in this chapter: play it or take the skill apart and build your own
Take an hour. Write down what you actually did last week, task by task, ten or fifteen lines. Not the job title. The tasks. Then mark each one:
- A. Could a machine carry this better than me, honestly?
- B. Did it give me energy or take it?
- C. Was it the reason anyone wanted me there?
Now the harder half, and it is the half almost nobody does. Show the list to somebody who knows your work and let them tell you which items you marked wrong. Column C is where people flatter themselves and where a colleague is worth more than an hour of thinking alone.
I have never met anyone who did both halves and came away with the same answer they walked in with.
That is the audit. The other half is the antidote, and it costs another part of an evening. Have one conversation with an AI this week with no objective at all, no deliverable, a question you cannot answer but it interests you, and follow it wherever it goes.