Which Jobs Go First: A Framework Rather Than a List

Lists of at-risk occupations age badly, because they're guesses about job titles when exposure operates at the level of tasks. A framework is more durable — you can apply it to any role, including ones that don't exist yet.

The four properties

For any task, score it on four axes. The scores predict exposure better than the job title does.

1. Input completeness. Is everything needed present in artifacts a machine can read? Or does it require context held by people — history, relationships, what someone meant rather than wrote?

2. Output verifiability. Can correctness be checked mechanically and quickly? Or does it need judgment, or take months to become apparent?

3. Consequence of error. Is a mistake cheap and reversible, or expensive and permanent?

4. Accountability requirement. Does someone need to be answerable for this, in a way that requires a party who bears consequences?

→ Complete inputs, verifiable output, cheap errors, no accountability requirement = highly exposed. Missing any one slows exposure substantially; missing several makes it unlikely for a long time.

Applying it

Data entry from documents. Complete inputs, verifiable against a schema, cheap errors, no accountability requirement. Highly exposed on all four — and this is the shape of work that goes first, in any domain.

Junior implementation from specifications. Complete inputs, verifiable by tests, moderately cheap errors (caught in review), accountability sits with the reviewer. Exposed on most axes, which matches what's being observed.

Diagnostic medicine. Inputs often incomplete (patient history, presentation, what they didn't mention), verification slow, errors expensive, accountability requirement absolute. Low exposure on three of four regardless of capability — and the assistive use is high-value precisely because the accountability stays human.

Contract review for standard clauses. Complete inputs, verifiable against a checklist, moderate error cost, accountability required but delegable to a reviewing lawyer. Mixed — which is why this is being partly automated with a human signing.

Front-line support for common issues. Complete inputs (usually), verifiable by resolution, cheap errors, low accountability. Exposed, which matches deployment reality.

Support for angry customers on complex problems. Incomplete inputs, unverifiable, expensive errors (the relationship), and someone has to own it. Not exposed, and this is where the remaining support work concentrates.

⚠️ Why title-level lists mislead

Every job is a bundle. Two people with the same title can have completely different exposure depending on which tasks they actually do.

A lawyer doing standard document review is exposed; a lawyer doing negotiation and judgment isn't. A support agent handling password resets is exposed; one handling escalations isn't. A developer implementing tickets is exposed; one deciding what to build isn't.

✅ Which means the useful exercise is scoring your own week, not looking up your title. That takes twenty minutes and gives you a real answer.

The second-order effects the framework misses

Being honest about limits — exposure isn't the whole story:

Bundle collapse. If the exposed 60% of a job is automated, the remaining 40% may not be a full-time role, and the job disappears even though the residual tasks weren't exposed.

Volume changes. Cheaper output may mean more demand, changing headcount independently of exposure. The framework says nothing about this.

Entry-level compression. The exposed tasks are often how people learn, so exposure affects the pipeline into a role separately from the role itself.

Organizational choices. Two companies facing identical exposure make different decisions based on their demand and their culture.

The takeaway

Score tasks on input completeness, output verifiability, error cost, and accountability requirement — not job titles, which bundle tasks with wildly different exposure. Apply it to your own week rather than to a list. Then account for what the framework misses: bundles can collapse when their exposed majority goes, demand can change independently, and the tasks that are most exposed are often the ones people learned on.