THE SHORT ANSWER
An AI-augmented professional redesigns selected work around clear inputs, reusable instructions, verification and measurement. They use domain expertise to judge outputs, document effective patterns and stop using AI where review cost or risk exceeds the benefit.
Frequency is not maturity
Occasional use produces isolated drafts. Augmented practice defines the recurring task, prepares reliable context, establishes review criteria, records exceptions and measures whether accepted output improves.
Build one reusable pattern
- Select a frequent, bounded task.
- Write the desired outcome and unacceptable errors.
- Create a reusable input template.
- Specify source and review requirements.
- Compare time, quality and rework with the previous method.
- Document when the workflow should not be used.
Productivity gains depend on people and task design
In the customer-support field study, gains varied substantially by prior experience. In the consultant experiment, results differed across task types. Together they support experimentation with measured workflows—not a promise that every professional becomes faster or better.
Evidence & context: National Bureau of Economic Research · Harvard Business School
Measure accepted outcomes
| Dimension | Question |
|---|---|
| Time | Did total time include preparation and review? |
| Quality | Did accepted work meet the same standard? |
| Rework | How often did errors return downstream? |
| Risk | Did exposure or consequence increase? |
| Learning | Can the person explain and improve the result? |
Evidence & context: GOV.UK Service Manual
Sources & further reading
- Generative AI at Work
National Bureau of Economic Research. A field study of 5,179 customer-support agents during a staggered AI-assistant rollout, later published in the Quarterly Journal of Economics. Results varied by experience and task and should not be generalized to all occupations.
- Navigating the Jagged Technological Frontier
Harvard Business School. A field experiment with 758 BCG consultants performing selected knowledge-work tasks. AI effects differed depending on whether a task was inside or outside the model's capability frontier; the sample and tasks limit broader inference.
- Measuring the benefits of your service
GOV.UK Service Manual. Guidance connecting time, cost and service measures to outcomes and benefits. It does not provide universal automation ROI assumptions.
Examples and exercises are illustrative unless attributed to a source. No independent expert review is claimed.
A correction, a counterexample or an experience worth sharing?
Join the conversation ↗