AI Can Teach Employees When It Acts as a Coach, Not a Substitute for Thinking

Harvard Business Review published an interview on July 23 with Angela Duckworth of the University of Pennsylvania. The psychologist connected her research on perseverance with the question of whether AI strengthens employees’ abilities or gradually weakens them. Perseverance, she says, is not short-term enthusiasm, but a long-term combination of interest and effort. “Enthusiasm is common, endurance is rare,” she said at the 2026 Leadership Summit. Duckworth draws on a survey conducted with the Walton Family Foundation and Gallup. A representative sample of about 2,000 Americans aged 18 to 28 responded at the end of October 2025. Three quarters had used AI during the previous month, most often as a replacement for internet search or to increase productivity. At the same time, young people expressed concern that a convenient tool might reduce their long-term independence and ability to think. A key experiment was led by then doctoral student Benjamin Lira. Participants had to revise a poor cover letter. One group worked alone, another could use AI, and a control group did not practise the task. Researchers expected the tool to improve the immediate output but weaken the underlying skill. They found the opposite: AI users wrote better without assistance both one day and one week later than people who had practised alone. The explanation lies in the quality of the learning environment. Participants using the tool made fewer keystrokes and spent less time on the task, so they exerted less effort. At the same time, they immediately saw a better example of structure and phrasing. When a system performs an activity better than a beginner and its output can be observed, it may act like a coach: explaining, demonstrating a model, and enabling imitation. One experiment does not prove the same effect across all occupations and tasks. Duckworth therefore does not deny the risk of declining skills. “We can become wiser, and we can also become less capable,” she warned. If people permanently hand all cognitive work to a system, they stop exercising judgement much like an unused muscle. Managers must distinguish tasks with little learning value, such as mechanical formatting, from work through which employees build professional judgement, argumentation, or the ability to detect an error. The implication is a different design for corporate training. It is not enough to show an employee how to write a request. The employee should first predict the result, compare it with the system’s response, mark errors, and explain what should change. For key skills, an unaided task should follow so that the company can verify genuine learning. Performance can be measured by immediate speed and by later independent results; minutes saved alone may conceal a decline in capability. According to Duckworth, the highest-level objective should remain stable while lower-level methods can change flexibly. A company can use AI for routine tasks while protecting activities that develop future experts. This directly affects entry-level roles: if the system takes over all basic work, the company must create another route to experience. AI becomes a coach only when a person actively studies its example. Terms explained Perseverance: A long-term combination of interest and effort in pursuing a stable goal despite obstacles. Control group: Participants in an experiment who do not receive the tested intervention and provide a basis for comparison. Learning environment: Conditions, examples, feedback, and procedures that influence the acquisition of a new ability. Cognitive offloading: Handing a task to a system so that the person no longer performs a substantial part of the reasoning.

AI Can Teach Employees When It Acts as a Coach, Not a Substitute for Thinking

The full article is available to users registered at Hard Skills