How to AI-proof your job

Good afternoon. In today’s letter I’m writing about some new studies that are starting to offer a better sense of which jobs are most vulnerable to AI, and how people can think about AI-proofing their careers. 

Plus: The U.S. blockades the Strait of Hormuz, rents are collapsing, parents spending thousands on career coaching for their kids, and Ray Dalio’s pessimism.

As always, send me an email if you have any comments, questions, or thoughts to share.

— Taylor Scollon

I have mostly tried to stay away from making predictions about how AI will impact the job market on a macro level because, while you’ll hear people in the industry making very confident calls about this (Dario Amodei, Geoffrey Hinton), there’s really not yet much data to support any particular view. 

We are, however, starting to get some solid evidence on how AI is impacting work at a more micro level, and one thing that is emerging is that AI’s effects are best understood at the level of tasks — all the things that a worker does that together constitute a job.

One paper from a couple of weeks ago breaks jobs down into those with “weakly bundled” tasks, where a job mostly consists of standalone tasks that could be easily assigned to someone else, and “strongly bundled” tasks, where you want the same person doing a variety of different tasks.

The most obvious example of a job with weakly bundled tasks is an entry-level coder, whose only job is to complete code-writing tasks assigned to them by someone more senior. The bundle of tasks this person is responsible for could be passed off to someone else — or to an AI tool — without much trouble. 

A senior software engineer, however, is a job with tightly bundled tasks. They probably will write code, but they are also responsible for reviewing other people’s code, taking business requirements and translating that into code, interfacing with other non-technical workers, managing teams, and so on.

By this account, AI is bad news for someone in a weakly bundled job, as most of their tasks can be handed off without too much friction. Someone in a tightly bundled job, on the other hand, may actually benefit from AI, as their task bundle as a whole cannot be done by the AI system. The effect is to free up more of their time for the human-critical tasks, which become more valuable.

I like this framework because it helps us think more clearly about what sort of jobs and careers will be endangered by AI. If you can break down a job into tasks, and at least one of those tasks is still human-critical, then stronger AI may be a net benefit for your career. If none of the tasks in a job bundle are human-critical, then that is a more at-risk job. 

This seems consistent with some of the early evidence about AI’s impact on jobs, like the oft-cited example of increased hiring for radiologist positions despite more use of AI in the field. That’s a tightly bundled job with a number of tasks that are a) best done together by a single person, and b) can’t all be done by AI (yet).

I say “yet” because there’s always the possibility that more advanced AI could flip a job that was previously complemented by AI into one for which AI is a substitute. I don’t have a strong view on how capable AI will become, so predicting which tasks are most sheltered from this risk is difficult. It’s also the case that AI tools themselves have “jagged” capabilities, in that they excel in certain tasks humans find challenging, but fail in ways that we find trivial — and the shape of this jaggedness appears to change over time.

Given all that, the most practical thing you can do as a worker, in my view, is to add tasks to your job bundle that AI tools seem to struggle with the most: coordinating people, managing clients, sales, non-technical writing, and design judgement are a few that come to mind, but there are certainly many more.

In this Smart Money Q&A with Dimitri Busevs, President & CEO of RBC Direct Investing, we ask: 

Read the full Smart Money Q&A on our website.