Look for resilient work—not “AI-proof” jobs.

Occupations change as tools, regulation, customer expectations and business models change. A label applied to a job title hides the mix of tasks inside it and creates false certainty.

The more useful question is which responsibilities are addressable, which still require a person and which could improve when a person uses AI well.

Resilience tends to increase when good work depends on changing context, trust, embodied action, accountable decisions or requirements that cannot be reduced to an easily checked digital output.

These are prompts for investigation, not guarantees about demand or employment.

A clinician may use AI to summarize notes while retaining responsibility for diagnosis. A developer may generate routine code while owning architecture and review.

A customer-service worker may draft responses while handling exceptions and emotional conversations.

High addressability in one part of a workflow does not prove that the entire role disappears.

Some work is both human-critical and suitable for assistance.

Look for a bounded task where a tool can reduce preparation or repetition while a knowledgeable person defines quality, checks output and remains accountable.

Record the corrections and failure modes; a polished demonstration is not evidence of a dependable workflow.

When comparing roles, inspect recurring tasks, entry pathways, regulation, physical requirements, customer contact and how junior staff learn.

Use current occupational data and real job descriptions.

Avoid switching careers solely because a list calls one job safe; the local labour market and your ability to enter and perform the work still matter.