The doorman fallacy in the age of AI

When the door opens itself, what still needs doing?

A doorman just opens the door, right? Replace the doorman with an automatic sliding door and the door still opens plus we save some money. Great, but who greets the regulars, watches for trouble, or finds you a taxi in the rain? Rory Sutherland calls this the doorman fallacy: when we reduce a complex job to its most visible and automatable task, we miss the full scope of the role with all other intangible aspects.

This is a useful way to think about jobs affected by AI. We have to go beyond thinking of mathematicians as just proving theorems, developers as just writing code, or doormen as just holding the door.

BEFORE Welcome back Adrian! Need a taxi? HOTEL After: the automatic door An automatic hotel door slides open. The same guest stands alone in the rain and asks who can call a taxi. AFTER “OPTIMIZATION” Okay, but can you call me a taxi? HOTEL whoosh

We automated the door. We forgot the doorman.

Before: the product owner and senior developer A product owner and senior developer decide what to build, what to skip, and how to keep the product coherent. Their computer displays Homer Simpson’s normal pink car. BEFORE Build this. Skip that. Keep it one product. After: every request gets built Every request has been implemented. The computer displays the supplied image of Homer Simpson’s extravagant car, while a colleague asks what the product has become. AFTER “OPTIMIZATION” We built every request. What is this thing? AI

We automated code. We forgot product management.

A senior developer writes code (the equivalent of holding the door), but also works with stakeholders to decide what to build/not to build, how to make the product intuitive and coherent, decide on the long-term architecture, hire, onboard, and coach people, and much more.

Asking the right questions will become more important across all disciplines that are impacted by AI. From a compute and context perspective, humans are very limited compared to AI, but we have intuition, ideas, and can connect the dots between domains.

AI might eventually take on some of those other responsibilities too, but we should ask ourselves: if the door opens itself, what still needs doing?

Inspired by Lex Fridman’s conversation with Ankur Moitra, Steven Brunton, and Alex Kontorovich on mathematics in the age of AI.