I'm trying to point at jobs that can "in theory" be done by a sufficiently good multi-modal LLM. If the job requires only the production of digital artifacts, it counts. On the other hand I'm excluding:

• Jobs that require physical interaction with the world

• Jobs that depend on significant interaction with humans external to the organization

In other words: I'm not asking when robotics will be solved, and I'm not asking when humans will be comfortable interacting with an AI. If a job is mostly/entirely "internal" in the sense that if it was replaced by an AI only people in that org would notice, it is included. If a job could be done by an AI pretending to be a human (e.g. a remote-only salesperson), it is included. Feel free to ask for clarification about particular categories of work.

The work must be fully automated. For example the current paradigm where senior engineers mostly coach a team of agents to do the coding they used to do directly does not count.

The work does not need to be in fact automated, only automatable.

Almost all: if there are random unimportant and obscure holdouts that are impossible to automate for some strange reason, question still resolves YES.

Market resolves to my subjective impression. For calibration I do not currently believe that almost any job is fully automatable, but I do think that will change within the next 1-2 years.