Comments
48
I hate clickbaitish titles as well, but the purpose of this post is to directly answer these two questions. If you’re in a hurry, jump to “Finding Out For Yourself.”
In early 2024, I was contracting for RLHF, developing small programs in 30 to 90-minute sessions to train AI in coding. It took me about 40 days to realize that the profession I love—and which I’ve been involved in for over a quarter of a century—is over. Or, to be more precise, it will be over “pretty soon”. I quit immediately (“I’m not digging my own grave”), and I was terrified.
In early 2025, I took the AI Safety, Ethics & Society course by the Center for AI Safety. I was still deeply disturbed, wondering: how much time do I have left? So I came up with the idea of doing project work about forecasting job market transformations. Luckily my mentor, Aisha, pointed out that I should narrow my scope for a four-week project. So I developed a forecasting methodology—which, actually, is “just” a rational estimation methodology. (The thing is: I’m Hungarian, and in the semantic space of the Hungarian language Forecasting and Estimation are extremely close. I didn’t know at the time that it’s not so in English).
Analytically and structurally thinking through the forces affecting the risks of losing my job helped me a lot with my anxiety. It helped me see a horizon, and also to see the directions where land may still be abundant ahead of me. The feedback from our learning group was positive, but at the time I was too busy with my day-to-day work.
In early 2026, after quitting my job to pivot my career toward AI Safety (the AISES course was quite effective), I thought my methodology could be beneficial and calming to others. I’ve turned my estimation method into a short paper. With it, one can create personal, structured job security estimations like this (construction worker), this (software developer), and this (dentist). The method is simple at its core: it decomposes occupations into narrower, key human capabilities and allows the user to estimate their "human advantage" against AI in those specific domains. It’s just likely easier to make better estimations in narrow domains, compared to broad ones.
I’ve also created an AI prompt from the estimation method. If you have access to a reasoning model, you can apply it to your specific circumstances in about five minutes.
Here’s what you can do: copy-paste, or download and attach this markdown describing the method into an AI chat and ask at the end: “Apply this methodology to the profession of [YOUR PROFESSION HERE]!” You can extend the prompt with some specifics and context about your situation — doing so improves the reasoning.
The five-minute result is usually already interesting, but you can iterate on that: argue with the AI on its takes where you disagree. After only a few iterations you’ll see a reasoning about the prospects of your job security that you’ll likely agree with — because it will be based on your own strong beliefs.
(In case of “emergency”, just continue: play around with the model to find paths toward more job security. It can guide you on how to reshape your job description to include more AI-resistant components.)
Go, try it out now!
(I've also written a longer article on LessWrong, detailing the reasoning process.)
I see a lot of people who are anxious about losing their jobs to AI. I also see a tremendous amount of denial and wishful thinking (especially among software engineers, but I’m biased because that’s my information bubble, and software engineering also seems to be affected in the front line). I think reasoning rationally and structurally about risks is a superior method of coping, so I think this method is genuinely useful.
I’m also very interested in your feedback! What do you think about the methodology? Was this useful to you?
