Use task-level questions to investigate where AI fits into team workflows. Examples are editorial prompts, not measured exposure or a forecast.
What this tool is for
Start with a specific task, then ask what is automated, who reviews the result, and who remains accountable for decisions.
Method and limits
The examples are not a role-level score, an exposure measurement, or a prediction about a person's employment.
Questions worth checking
- Does this page predict AI job loss?
- No. It offers editorial task-level questions for investigating team workflows; it does not measure role exposure or forecast employment outcomes.
- What can I verify with my team?
- Ask what work is automated, who checks the result, what source material supports it, and who remains accountable for the decision.
Start with a task, not a job title
Write down one recurring piece of work from intake through delivery: its inputs, transformation, review, hand-off, and consequence. Ask which step a tool can draft, search, classify, or repeat, then ask what context must still be supplied by a person. A job title can bundle routine production with negotiation, exception handling, customer trust, and accountable decisions. Those tasks may change at different speeds inside one team. This worksheet is a set of investigation prompts, not a measure of exposure, a market forecast, or an estimate of employment risk.
Illustrative scenario: trace an output to its owner
Illustrative scenario: a team uses a language model to prepare a first draft of a customer update. Map who provides the source facts, who checks claims, who removes confidential information, who approves the tone, and who handles an incorrect statement. Record whether the tool changes the amount of work, its distribution, or the quality-control burden. A faster draft can move effort toward verification rather than remove it. Do not submit restricted or personal material to a system unless your employer has approved that use. The example describes a workflow question, not a prediction about a particular occupation.
Make a practical learning test
Choose one permitted task and compare a normal process with an assisted process using the same quality standard. Track time, correction types, missing context, and who accepts responsibility. Keep the source and an unassisted baseline where policy allows. Ask your manager how the result will be reviewed and whether the tool is approved for that data. A useful skill demonstration shows judgement: what you rejected, what you checked, and why the final output is fit for its audience. Tool familiarity without a real workflow or review boundary is weak evidence of contribution.
Questions for a team conversation
- Does this worksheet say whether AI will replace me?
- No. It helps identify task changes and accountability questions; it cannot forecast adoption, staffing, or future demand.
- What should I ask before using a tool?
- Clarify approved products, data handling, review ownership, permitted sources, escalation routes, and what happens when an output is wrong.
Map exceptions and hand-offs
Routine examples rarely reveal the hardest part of a workflow. List the exceptions: incomplete records, contradictory instructions, unusual customer needs, policy conflicts, or a failed upstream system. Ask who detects each exception, who can pause the process, and where the case goes next. A tool may produce plausible text while omitting the condition that changes the correct response. Reviewing edge cases helps distinguish a draft generator from a system that can safely complete a task. Keep a record of the review criteria and escalation path rather than relying on a general instruction to check the output.
Protect data and provenance
Before testing a tool, identify the data classification and the system the employer permits. Do not paste customer records, employee details, source code, confidential strategy, or unpublished financial information into an unapproved service. Ask whether prompts or outputs are retained, who can access them, and whether generated material can be traced to source documents. For a public-facing answer, preserve citations and verify that they support the actual claim. A polished response is not evidence of correctness. If the source cannot be checked, mark the claim unresolved and keep a human owner responsible for the final decision.
Describe your contribution precisely
If you use an assistant in a work sample, explain the task, the permitted inputs, the parts you reviewed, and the changes you made. Show how you noticed a wrong assumption or improved a weak result. Do not claim that the tool created expertise or that a faster draft proves a business outcome. Ask your manager whether disclosure is required and follow team policy. A useful portfolio case distinguishes your judgement from automated output and respects confidentiality. This gives a hiring conversation something concrete to examine: problem framing, quality checks, exception handling, and accountability rather than a list of software names.