Skip to content

AI-era career ratings

Rate the career paths you’re weighing — like stocks.

Picking a career is the biggest bet most of us ever make, and most of the advice is vibes. CareerStar treats it like an analyst treats a stock: every U.S. occupation gets one 0–100 score — real growth and pay data, discounted by how exposed the work is to AI, blended with how well it fits you. No horoscope. Math you can check.

730 careers rated· built on real U.S. government data· every score is explained, not guessed

Tested against a real decade, not vibes. I scored 2014’s careers using only what was knowable in 2014, then checked what actually happened by 2024. Careers the model scored higher really did fare better on average — and it caught 48% of the careers that went on to shrink, where random guessing would catch 33%. Where it missed, I say so. See the test →

01 · The process

How it works

1

Name your paths

Careers, whole fields, or just your interests — start anywhere, even from “I don’t know yet.”

2

The math rates each one

How much a career pays off (growth + pay), marked down by how much of the work AI could take over, blended with how well it fits you — the same way an investor judges a stock: reward, discounted by risk. An AI writes the plain-English why, never the number.

3

Read the verdict — and the receipts

A ranked comparison with star ratings, the strongest case for and against every career, and the exact math behind each score, one click away.

02 · Second opinions, built in

Five judges, not one oracle

Any single formula is one opinion about how much AI risk should count. So every comparison is scored by five rival models — from “ignore AI entirely” to “safety is everything.” When they agree, that’s conviction. When they don’t, that’s a real finding about your choice — and you can pick which judge scores your ranking, with each one’s strengths and blind spots stated up front.

Standard
the balanced judge — AI risk discounts the reward
+ The balanced read — and the only model here that's been tested against a real decade of history (2014→2024).
− Its weights are judgment calls — no dataset proves them right — which is exactly why the four rivals exist.
Growth optimist
the optimist — growth and pay only, ignores AI on purpose
+ The purest read of the official forecasts — and in the 2014→2024 test, the raw forecast was the single best predictor.
− Blind to AI by design. A heavily exposed career with good growth numbers scores well right up until it doesn't.
Defensive
the safety-first judge — a career's AI shelter matters most
+ Built for the bad scenario — the only judge that scores how sheltered a career is from AI, so it catches safety the others ignore.
− Punishes exposed-but-booming fields hard: it will underrate software developers in every future where AI disappoints.
Efficiency
the efficiency judge — reward per unit of risk
+ Flags high-reward careers whose risk is quietly outsized — something the other judges let slide.
− Ratios flatter the timid — a low-reward, low-risk career can score 'efficient' while going nowhere.
Plain average
the plain average — the baseline the others must beat
+ No opinions to be wrong about — just the bar every fancier model has to clear.
− Counts every ingredient equally, which nobody actually believes — including it.

03 · The receipts

Why you can check it

  • No black box. The same inputs always produce the same score — no AI guessing, no randomness — and every formula is published in full on the how-scores-work page, including why each piece was chosen over its alternatives.
  • Tested against history, misses named. The model rated the careers of 2014 and was graded on what actually happened by 2024 — including the careers it got wrong, listed by name.
  • Stress-tested. Every comparison is re-scored 729 extra times with the model’s settings nudged up and down, and again by five rival formulas. A ranking that flips when the settings wiggle is flagged as a close call, not sold as a verdict.
  • Open data. Every rating is in one downloadable CSV, and the code is on GitHub. If a number looks wrong, audit it.

Ninety seconds from “no idea” to a ranked answer.

Free, no sign-up, nothing stored. The worst case is you disagree with the math — and you can see all of it.

Rate my paths →