what jobs are in danger due to ai?
the exposed work is screen-based, produces a document or a decision, and needs no licensed signature: first-draft copywriting, basic translation, tier-one support, data entry, junior paralegal review, and routine bookkeeping. entry-level rungs are shrinking fastest, which is the more damaging effect.
the exposure test
A task is exposed when three things are true at once: it happens entirely on a screen, its output is text, code, an image, or a routine decision, and no licensed human has to sign for it. Score any job against that and you get a better answer than any ranked list.
Highest exposure by that test:
| work | why it is exposed | what actually happens |
|---|---|---|
| first-draft copywriting | text out, no signature | volume moves to generation, editors remain |
| basic translation | text in, text out | post-editing replaces translation |
| tier-one support | scripted, high volume | deflection rises, escalation roles grow |
| data entry and extraction | structured output from documents | close to fully automated already |
| junior paralegal review | pattern matching over documents | the pyramid narrows at the bottom |
| routine bookkeeping | rule-based reconciliation | fewer people, more exception handling |
the entry-level problem
This is the part that deserves more attention than the occupation lists. Almost everything above is how people used to enter a profession. Junior paralegals became senior ones. Tier-one support became product specialists. First-draft writers became editors.
Automating the bottom rung does not just remove those jobs. It removes the training path to the jobs above them, and there is currently no replacement for it. A firm that hires no juniors this year has a senior shortage in seven, and almost nobody is planning for that.
what is less exposed than the headlines suggest
Three categories keep getting listed and keep not happening:
- Software engineering. Code generation is genuinely strong, and the job was never mostly typing. Specification, review, and system design have grown. Entry-level hiring has fallen, which is the same rung problem again.
- Radiology. The canonical prediction, made in 2016, and radiology employment grew. Reading scans got faster, demand for scans rose, and the accountability stayed human.
- Accounting. Compliance and exception handling absorbed the freed capacity.
The pattern is consistent enough to be a rule: where demand is elastic, faster output means more work done rather than fewer workers.
what to do about it
For individuals, the durable position is on the accountability side of the output: reviewing, specifying, and being answerable, plus depth in a domain that a general model does not know. For managers, the honest move is to keep hiring juniors and change what they do, because the alternative is a hole you cannot fill later.
I build software that runs teams of agents, and the constraint I see is never how much they can produce. It is how fast a competent human can check it. That job is growing quickly and nobody is training for it. Which 5 jobs will survive ai covers the other side of the same question.
related questions
what workers is ai replacing?
so far: data entry, basic translation, first-draft copy, tier-one support, and junior document review. mostly tasks rather than people, and mostly at the entry level.
which 5 jobs will survive ai?
no list of five is authoritative. the roles that hold up share three traits: physical presence, legal accountability, or a relationship the work cannot be separated from.
what 3 jobs will not be replaced by ai?
skilled trades, hands-on healthcare, and roles carrying legal accountability. each survives for a different reason, and the reasons are more useful than the list.
what is the 30% rule for ai?
there is no single 30% rule. the phrase covers three different claims: an automation ceiling, a productivity gain, and a hallucination budget. only one of them is useful.
spin up your first room.
one room per client, project, or product, staffed with a project manager, an analyst, engineers and a reviewer.