The AI Productivity Paradox: Faster Results Could Come at the Cost of Human Motivation

Generative AI can dramatically accelerate productivity, but ADB research warns that diminishing returns may encourage workers to stop early, limiting final quality gains. For governments and businesses, capturing AI’s full value will require stronger incentives, skills development and work design that rewards excellence beyond “good enough.”

Generative artificial intelligence is emerging as a powerful productivity tool, particularly for economies and businesses facing shortages of specialized skills. But research from the Asian Development Bank (ADB), the University of Hong Kong and Arizona State University suggests that faster work does not automatically mean better final work. AI can help workers reach a good result much faster, but it can also reduce the incentive to keep refining that result, creating what the researchers describe as a potential "mediocrity trap."

The study examined 219 professional and student artists. Of them, 168 were given access to generative AI during their second illustration task, while 51 formed a control group. Around 79% of those given access actually used the technology. Each task could take up to 150 minutes, and researchers evaluated the work at 15-minute intervals to understand how AI changed both productivity and effort.

AI Delivers a Powerful Early Productivity Boost

The strongest benefit appeared during the early stages of work. AI-assisted creators reached an average quality score of around 50 in only 15 minutes. Without AI, reaching roughly the same quality level took about 60 minutes.

This productivity jump could have major economic implications. Developing countries with shortages of skilled workers could use AI to increase output in creative and knowledge-intensive industries. Companies could complete routine assignments faster, reduce production costs and allow employees to handle larger workloads.

However, the advantage weakened as people continued working. AI significantly improved quality during the first 60 minutes, but the effect declined afterward and became statistically insignificant after 90 minutes. Among creators working for the full 150 minutes, the difference between AI-assisted and conventional work was no longer statistically significant.

This means policymakers should be cautious about measuring AI success only through time saved. Productivity indicators should also examine final quality, innovation and long-term development of worker skills.

When Faster Work Reduces the Incentive to Improve

AI access reduced average working time by approximately 24 minutes from a baseline of 122 minutes. Overall, 77% of creators spent less time on their AI-assisted assignment.

A particularly important change occurred around the 60-minute mark. Without AI, many participants continued toward the 150-minute limit. With AI, a new concentration of creators stopped at around 60 minutes, the stage at which additional improvements had become relatively small.

Researchers found that a 10-point decline in quality improvement during the previous 15 minutes was associated with a three-percentage-point increase in the probability of stopping.

The result highlights an important distinction between productivity and motivation. AI makes it easier to reach an acceptable standard, but workers may decide that spending considerably more time chasing relatively small improvements is no longer worthwhile.

That trade-off affected final results. Around 62% of creators achieved higher-quality output with AI, but 38% experienced lower final quality. Most of those whose quality declined also reduced their working time. Overall, the improvement in final quality was only 0.11 standard deviations and was statistically insignificant.

Developing Economies Face a Skills Opportunity and a Risk

For governments, particularly in developing economies, AI could provide a new way to address shortages of specialized skills. The research indicates that AI may reduce differences linked to technical implementation capabilities. Quality gains were directionally larger among lower-performing creators, although differences across skill groups were not statistically significant.

That creates opportunities for countries seeking to expand digital services, creative industries and other knowledge-intensive activities. Workers may be able to perform tasks that previously required stronger technical capabilities, potentially lowering barriers to participation.

But there is a longer-term development risk. Difficult stages of work are also where professionals build experience, judgment and tacit knowledge. If AI encourages workers to stop once output becomes "good enough," greater short-term efficiency could come at the expense of learning-by-doing and expertise development.

Governments may therefore need to redesign education and workforce policies around skills that become more valuable in an AI-intensive economy, including creativity, critical thinking, problem definition, verification and advanced refinement.

From AI Adoption to Smarter Work Design

For international development partners, simply financing AI tools and digital infrastructure may not be enough. AI programmes could combine technology access with workforce training, institutional capacity building and systems for measuring whether AI actually improves quality and skills.

Private companies face a similar challenge. Businesses producing standardized content may benefit significantly from faster AI-assisted workflows. Firms competing through premium quality, however, may need stronger incentives and quality-control systems that encourage employees to continue refining work after AI has generated an acceptable first result.

The research suggests that organizations may need to rethink performance management as AI spreads. Instead of rewarding workers mainly for completing tasks quickly, employers could incorporate quality benchmarks, human review, experimentation and continued skills development.

For policymakers, development institutions and businesses, the central lesson is that AI adoption should be treated as an organizational and human-capital transformation rather than simply a technology upgrade. Generative AI can compress substantial amounts of work into minutes, but sustaining excellence still requires human effort. The policy challenge is therefore to capture AI's speed and productivity benefits without allowing "good enough" output to become the new ceiling for skills, innovation and competitiveness.