We Ran 142 Jobs Through the "Can AI Do This" Test. Here’s What Survived.
Every job in this catalog had to pass the same three-part test before it made the cut. Here’s the test, and what it actually catches that other "AI-proof" lists miss.
A job survives our test if it needs a body physically present, judgment that changes with conditions on the ground, and a failure mode that costs more than a support ticket. Jobs that look physical but are actually repetitive and predictable fail the test just as fast as desk jobs do. Here’s how it actually plays out across the catalog.
Every "AI-proof jobs" list right now is a guess dressed up as an analysis. We built an actual filter and ran the entire 142-job catalog through it before anything got published. Here's the test, and a few results that surprised us.
The Three-Part Test
- Does the environment change every time? A job where conditions repeat (same input, same output, every time) is automatable no matter how physical it looks. A job where every instance is genuinely different (a different snowpack, a different damaged ship, a different horse's hoof) is not.
- Is the failure mode a body, not a ticket? If getting it wrong means someone gets hurt, you need a human who can be held responsible and who can improvise mid-task. Software failure modes get patched. This kind doesn't.
- Does the physical task itself require a body on site? Not "could a robot theoretically do this in a lab," but "is anyone actually deploying that today, at this cost, in this environment." Most aren't, and won't be for a long time.
What Passed Clean
Commercial Diver, Farrier, and Smoke Jumper pass all three without argument. Different conditions every time, real physical consequences, and a physical task nobody is remotely automating. These are the clearest cases in the catalog.
What Passed for a Less Obvious Reason
Fish Monger (Specialty Auction) isn't dangerous and isn't extreme, but it passes because the actual skill, judging a bluefin tuna's quality by look and touch in seconds, at auction, with real money on the line, is a judgment call built on years of sensory calibration that doesn't reduce to a formula. That's a different flavor of "AI can't do this" than jumping out of a plane, and it's worth naming separately.
A job where you do the same motion on the same object under the same conditions, over and over, is closer to automatable than a desk job that requires constant judgment calls under changing conditions. "I work with my hands" is not the same claim as "what I do can't be automated." Plenty of repetitive manual labor is more exposed to automation than people assume. The test is unpredictability plus consequence, not "am I sitting down."
Why This Matters More Than a Vibe
Every job on this site passed the same test before it was added. That's the difference between a catalog built on a real filter and a listicle built on whatever sounded good that week.
Not Sure Which of These Fits You?
Take the 6-question quiz. It ranks all 142 jobs in the catalog by how well they match what you actually want.
Find My Path →
