Better Technology, Weaker Motivation: How Generative AI Can Undermine Excellence at Work

Generative AI can help workers reach acceptable quality four times faster, but reduced effort may offset its productivity gains and lower final output for some users. Governments, development partners and businesses must combine AI adoption with training, quality controls and incentives that reward sustained effort, originality and excellence.

Generative artificial intelligence can help people produce acceptable work much faster, but it may also weaken the motivation needed to achieve excellence. Research by the Asian Development Bank's Economic Research and Development Impact Department, the University of Hong Kong and Arizona State University shows that AI's productivity gains can be reduced when workers stop earlier because additional effort appears less rewarding.

The findings matter for governments, development partners and businesses investing in AI to address skill shortages, lower production costs and expand digital services. The study suggests that providing advanced technology is not enough. AI adoption must be supported by better training, incentives, quality standards and workplace design.

AI Delivers Speed, but Excellence Still Takes Time

The researchers conducted a randomized experiment involving 219 professional and student artists. All participants completed an illustration without AI. During the second assignment, 168 received access to a text-to-image system, while a control group of 51 continued without it. Around 79% of those offered the technology chose to use it.

Researchers recorded the production process and examined illustrations every 15 minutes. Each image was assessed by at least five university professors and industry professionals.

The productivity benefit appeared immediately. AI-assisted creators reached an average quality score of 50 in just 15 minutes. Artists working without AI needed around 60 minutes to reach the same level. The technology quickly generated images with composition, colour, lines and shading, reducing the burden of manual execution.

However, AI-assisted quality soon reached a plateau. Further prompting, editing and refinement produced progressively smaller improvements. Among participants who worked for the full 150-minute limit, the quality difference between AI-assisted and conventional work was no longer statistically significant.

The Hidden Cost of Stopping at "Good Enough"

Without AI, participants spent an average of 122 minutes completing an illustration. Access to AI reduced working time by approximately 24 minutes, or 0.63 standard deviations. The control group showed no similar decline.

AI also changed when people stopped working. Without the technology, many participants continued until the 150-minute limit. With AI, a new peak appeared at around 60 minutes—the point at which additional improvements had become very small.

Around 77% of AI-assisted creators reduced their working time, while 62% achieved better final quality. However, 38% produced lower-quality work despite having access to a more productive tool. About 32% both worked for less time and delivered poorer results, suggesting that they accepted lower quality in exchange for faster completion.

Average final quality increased by only 0.11 standard deviations, and the improvement was not statistically significant. AI therefore raised productivity at each comparable point in the process, but much of that benefit disappeared because creators stopped earlier.

A New Challenge for Skills and Development

The research suggests that AI could help developing economies compensate for shortages of designers, programmers, researchers and other specialized workers. It may enable less-experienced employees to reach competent performance faster and help smaller businesses enter digital and creative markets.

AI also appeared to reduce differences linked to technical execution. Lower-performing participants recorded directionally larger gains, although differences between skill groups were not statistically significant. As AI took over more manual production, creative ideas, judgement and conceptual direction became more important.

This shift creates a development risk. Workers build expertise by solving difficult problems, correcting mistakes and refining work over time. If AI encourages them to stop after producing an acceptable result, they may lose opportunities to develop the deeper knowledge required for innovation and higher-value employment.

Governments should therefore measure AI programmes through output quality, accuracy, originality and skills development—not only the number of users or hours saved. Schools and training institutions should teach people to evaluate AI outputs, detect errors, improve ideas and explain their decisions. Foundational knowledge must remain central because workers cannot supervise technology effectively without understanding the task.

Turning AI Productivity into Lasting Value

International development partners can support countries by financing digital infrastructure, teacher training, worker reskilling and independent assessments of AI programmes. Pilot projects should compare AI-assisted and conventional workflows across sectors such as education, healthcare, public administration and professional services.

Development institutions should also track whether AI improves employment, wages, service quality and domestic innovation. The study involved 219 illustrators, many of them young students, who used one image-generation tool and were limited each task to 150 minutes per task. Its precise numerical results should not automatically be applied to every industry.

For private companies, AI offers faster production, lower costs and support for less-experienced employees. The risks include overdependence on automated content, weaker employee engagement, reduced originality and erosion of specialist skills.

Businesses may increasingly have to choose between standardized "good enough" products and premium services requiring deeper human judgement. Companies pursuing high quality should reward accuracy, originality and sustained refinement. Firms focused on standardized production should redesign workflows so employees use AI for speed while concentrating human effort on verification, customer needs and risk management.

The main lesson is that AI adoption is not simply a technology purchase. It is a reform of work, incentives and skills. Without strong quality controls and rewards for sustained effort, generative AI may make competent work easier while making excellence increasingly difficult to maintain.