AI Job Survival · A Caterbum tool

How much of your job will AI do before you’re ready?

Paste a job description. Get a read on which parts of the work AI can already do, which parts it’ll reach in a few years, and which parts will still need a human in a decade.

First three analyses are free. About thirty seconds each. No signup to try.

Use it to answer one of these
For workers

“Is my job safe?”

Run your current job description. The result shows which responsibilities AI is already replacing — and which ones are still safely yours.

For job-hunters

“Will this job still exist?”

Run a posting before you apply. A role can pay well today and not exist in five years. The breakdown tells you which.

For students & parents

“Is this major worth it?”

Run the jobs graduates of a major typically take. See whether the degree leads to work that’ll still be there at graduation.

Why this is different

Most “AI exposure” tools give you one number for your whole job. That number is wrong.

Peer-reviewed analysis has found that leading AI-exposure scores disagree with each other by enormous margins — sometimes the measures are anti-correlated. The reason: the job isn’t the right unit of analysis. The taskswithin a job are, and they don’t all face the same risk.

This tool breaks a job into its underlying tasks and scores each one across five levels of complexity. Some tasks are codified and AI is replacing them now. Some require deep institutional knowledge AI hasn’t reached yet. Some are integrative work — holding multiple domains together at once — that current AI architectures are structurally bad at.

What you get back is a map of where in the job the risk lives, not a single panic number.

Try it

Paste a job description below.

Duties, qualifications, day-to-day responsibilities — the more detail, the better the read. Not stored. Not shared. Not used to train anything.

What you’ll get back

The result isn’t a score. It’s a map.

A single risk number tells you nothing useful. The output here is structured so you can act on it.

01

Distribution across L1–L5

The percentage of described tasks falling at each level.

02

Task-by-task placement

Which specific responsibilities sit at which level, and why.

03

AI exposure summary

Plain-language read of what AI is likely to replace in 2–5 years and what it isn’t.

04

What to develop

If the role is heavy on lower levels, which higher-level capabilities would shift the balance.

Reading the result

A high L1/L2 score is information, not a verdict.

The point of the tool isn’t to tell you a job is safe or doomed. It’s to show you where in the work the risk is — so you can do something about it.

If the result says 70% L1/L2

“The majority of described tasks are tasks AI can do soon. The role in its current form is at significant risk.”

What this doesn’t mean:That youare at risk. The tool reads the job description, not you. The same description can fit a junior person doing the work mechanically and a senior person who has stretched the role to include L3/L4 work that isn’t written down. If the gap between the description and what you actually do is large, the result is a signal to make that gap visible.

If the result says 50% L3, 30% L4

“The job will likely survive — and it will change.”

What this means:The L1/L2 parts will get automated, and the L3/L4 parts will become a larger share of the role. Whether that’s good or bad for you depends on whether you’ve been developing the L3/L4 capabilities or coasting on routine.

If the result says 60% L4, 20% L5

“This is the strongest signal of durability the tool can give.”

What this means:The role requires integrative judgment, tacit knowledge, often relational presence. It’s durable because it’s hard. The question to ask is whether you have those capabilities yet, or whether the role is currently being held together by someone else’s expertise that you’re meant to absorb over time.

What this tool can’t tell you

The limits.

The tool reads job descriptions, not jobs. A poorly written description produces a misleading result. A description that captures formal duties but misses the informal L3 work will underrate the role.

It predicts task-levelAI exposure, not employment outcomes. A job whose tasks are automatable can still survive for regulatory or relational reasons. A job whose tasks aren’t automatable can still get cut.

The framework itself is a proposal, not a settled scientific finding. It was built because no existing scale mapped work this way. There are four serious counterarguments worth knowing. If you find the framework wrong, that feedback makes it sharper.

Reference

The five levels.

A condensed version of the framework. The full essay is here if you want the long version.

L1
Single-domain, well-bounded. One domain, codified knowledge, clear inputs and outputs. Largely already automated.
L2
Multi-domain, sequential, low context. A clean pipeline of bounded steps. Orchestrator-plus-specialist architectures handle this well.
L3
Multi-domain, sequential, high context. Same pipeline shape, but requires institutional or relational knowledge not in any document.
L4
Multi-domain, simultaneous, integration-as-deliverable. Three or more domains held at once because the integration IS the deliverable. Resists decomposition.
L5
Multi-domain, with embodied or relational stakes. Level 4 plus embodied presence, live relational stakes, judgment that isn't fully articulable.