RedLectern topic
Workplace adoption · task automation · augmentation · productivity · labor-market evidence.
For: Workers · managers · policy readers.
After this page you can: Evaluate a workplace use case and interpret job-impact evidence.
Page reviewed
Part of: Artificial Intelligence › AI and society
AI at work is best assessed at the level of tasks and workflows. A role can include tasks that AI may assist with, tasks it may automate, and tasks that still depend on people. Occupational exposure describes potential effects on work; it does not tell you whether an employer will adopt a system or what will happen to employment, wages or demand.
To evaluate a use case, identify the task, the system’s role, the human checks, and the outcomes that matter: quality, time, errors, workload and worker experience. Then ask what evidence was collected, from whom, and under what conditions. Exposure estimates are signals about where work could change, not predictions of job displacement or productivity gains. For managers and policy readers, adoption choices and worker consultation matter alongside technical capability. For workers, the practical question is how a workflow changes the work people do and the outcomes they experience.
Read a short guide to what exposure indicators can—and cannot—tell you about work.
Start with: Workers’ exposure to AI: What indicators tell us – and what they don’t
Use survey evidence to consider worker experience, training and consultation.
Start with: The impact of AI on the workplace: Main findings from the OECD AI surveys of employers and workers
Start with the task, rather than treating a job title as one indivisible unit. Automation means a system performs a task with less direct human execution; augmentation means technology assists people as they perform tasks. Occupational exposure describes the potential for a technology to affect tasks in a job. These ideas describe different relationships between technology and work, and a job can contain tasks affected in different ways.
An exposure measure is not an employment forecast. It may help identify tasks worth examining, but it does not establish whether an employer will adopt AI, how a workflow will change, or what happens to employment, wages or demand. When evaluating a claim, ask which tasks were assessed and whether the evidence describes potential capability or observed workplace change.
Consult the task-based methodology behind one occupational exposure index.
Automation
Having a system perform tasks with reduced direct human execution.
Augmentation
Using technology to assist people in performing their tasks.
Occupational exposure
The extent to which a job’s tasks could be affected by a technology.
Choose a job you know. Name one task a system might perform with less direct human execution and one task where it might assist a person. What would you need to observe before calling either change an employment effect?
For example, a system might draft routine replies, while a worker reviews and adapts them for a specific case. To identify an employment effect, I would need evidence about adoption and changes in employment or work—not just an estimate that the task could be affected.
A workplace use case is more than a tool demonstration: describe the task, who uses the system, where a person checks or changes its output, and what happens when it is wrong. Compare the workflow before and after adoption using outcomes that matter for the work, such as quality, time, errors, workload and worker experience. Keep the conditions visible so that a result from one setting is not treated as a result for every job.
Workplace adoption also involves people and organizational choices. OECD surveys of employers and workers in manufacturing and finance across seven countries found that training and worker consultation were associated with better outcomes for workers. Treat those as considerations when assessing a deployment, and ask workers how the workflow affects their tasks and working conditions.
Read a country-specific examination of reported workplace performance and job quality.
A team proposes AI assistance for sorting incoming requests. Write one sentence naming the task, one measure of quality, one measure of worker experience, and one question to ask workers before rollout.
Example: “The system will sort incoming requests for staff review; we will track misrouted requests and resolution time, ask workers whether the workflow adds or removes useful discretion, and consult them about exceptions before rollout.”
Separate four kinds of evidence: estimates of which tasks could be affected, records of workplace adoption, measured or reported productivity, and observed labor-market outcomes such as employment or wages. They answer different questions. A task-based exposure index can flag where work might change, while a worker survey reports respondents’ experiences; neither alone shows what happened to employment across a labor market.
Check the study’s population, occupations, location, time period, and outcome before applying its result elsewhere. Exposure measures can rely on static task descriptions and may omit economic or institutional constraints on adoption. For a job-impact claim, look for evidence of actual adoption and labor-market developments alongside exposure estimates, and keep uncertainty about how workflows and demand may adjust.
Compare a global task-based exposure measure with observed employment evidence.
A report says an occupation has high AI exposure. List two additional kinds of evidence you would want before concluding that employment in that occupation changed.
I would want evidence that employers actually adopted the technology and observed labor-market data, such as employment or wage changes over time. I would also check whether the evidence covers the same occupations and period as the exposure estimate.
No recent updates are available.
Carl Benedikt Frey · 2019
A historical frame for asking how automation changes the distribution of economic and political power.
Thomas H. Davenport and Steven M. Miller · 2022
Case studies show how AI systems fit into real work settings, including the tasks people do with them. Treat these as cases from the book’s publication period, not evidence about today’s models.
Erik Brynjolfsson and Andrew McAfee · 2014
A broad account of how digital technology can affect productivity and work. Its 2014 perspective helps frame labor-market questions, while its examples predate current workplace AI deployments.