Focus on workflows, not extinction lists.

A new tool may change who drafts, who checks, how quickly an output is produced and which exceptions reach a specialist.

That can alter workload, staffing and entry-level learning even when the occupation remains. Track the sequence of work rather than treating adoption as a yes-or-no event.

Tasks involving language, code or structured data are more addressable when inputs repeat, outputs are clearly specified and a knowledgeable person can verify the result cheaply.

Addressability falls when context is missing, errors are consequential or the work depends on physical conditions, trust or accountable discretion.

Review, exception handling, stakeholder explanation, data governance and accountability can become a larger share of a role after routine production becomes faster.

This is not automatically positive: verification can be demanding, and organisations may introduce tools without enough training or clear ownership.

Useful signals include approved-tool policies, changed quality checks, new performance expectations, altered hiring, redesigned junior work and repeated use on production tasks.

Product launches and viral demonstrations show capability under selected conditions; they do not prove adoption, reliability or employment effects in your context.

Choose a low-risk task, record the current time and error rate, define the human check and test only with approved data and tools. Compare the full workflow, including corrections.

Keep the experiment when it improves a result without obscuring responsibility; stop when quality, privacy or ownership becomes unclear.