When AI does more of the thinking, do ideas get worse? Four experiments found a workflow effect
Across four randomized online experiments with 728 participants, high-assistance or high-delegation workflows reduced reported planning and checking. In one 176-person experiment, blind experts scored proposals 3.95 versus 4.78 out of seven—but the workflows differed in more than delegation alone.
The short version
- What happened
- Four randomized online experiments recruited 150, 162, 176 and 240 valid participants—728 in total—to complete short product, service, public-space or employee-benefit proposal tasks.
- Who it affects
- For managers, teachers and professional teams, the study supports testing workflows that require a human problem frame, alternatives and a final evidence check before accepting fluent AI output—not banning AI or assuming every use weakens thinking.
- What remains unproven
- Four separate experiments did not estimate the full proposed pathway in one sample
Keep in mind
Four separate experiments did not estimate the full proposed pathway in one sample
Online Credamo samples and short simulated innovation tasks
Compound workflow manipulation changed response detail, effort instructions and responsibility
Metacognitive disengagement and offloading were self-reported
Expert ratings were reliable but partly subjective
No workplace productivity, delivered-project, longitudinal or skill-retention outcomes
AI literacy was measured, not experimentally trained
Source analysis · 1 linked source. Primary analysed source date: 28 October 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 4 min The evidence, methods, implications and limitations
More complete AI help changed how participants approached the task
A new peer-reviewed study tested a concern familiar to many knowledge workers: when an AI supplies a nearly complete solution, do people stop planning, checking and reflecting? Across four separate randomized online experiments, 728 participants used researcher-configured AI assistants for short innovation tasks. High-assistance groups received solution-oriented responses; low-assistance groups received brief factual help and were expected to develop the core proposal themselves. The researchers measured what they call task-level metacognitive disengagement with a new nine-item self-report scale covering reduced planning, weaker monitoring and absent reflection.
The first experiment, with 150 participants designing a smart-home proposal, found higher disengagement in the high-assistance group, a standardized difference of d=0.67. A 162-person service-design experiment linked that difference partly to reported cognitive offloading. These results concern an immediate task response, not a permanent loss of skill. The paper validates the scale across the four samples, but it does not follow participants to see whether the effect persists or accumulates.[1]
Blind expert ratings found a quality gap, with an important causal limit
The clearest performance test involved 176 participants designing a public-space renewal proposal. Three experts in urban planning or community design, blind to the assigned condition, rated novelty, usefulness and developability on seven-point scales. High-delegation proposals averaged 3.95, against 4.78 for low delegation; d=0.75. The high-delegation group also reported more cognitive offloading and disengagement. A fourth, 240-person experiment found the negative indirect association was stronger for a more complex employee-benefits task and weaker among people reporting higher AI literacy.
The comparison should be read as a workflow effect, not proof that AI assistance by itself makes ideas worse. High-delegation participants received more detailed responses, were encouraged to rely on the tool and held less responsibility for core content. The low-delegation group received only factual support. Those bundled differences prevent the study from isolating delegation alone. All tasks were short online simulations, expert scores remain partly subjective, and no field team, delivered project, productivity measure or later skill test was observed.[1]
The distinction that matters
What changed between the two workflows
The experiment compared packages of instructions and AI support, not simply AI present versus AI absent.
High delegation
Detailed AI responses and encouragement to offload core work
Lower expert-rated proposal quality in Experiment 3
Low delegation
Factual AI help while the participant developed core content
Higher average novelty, usefulness and developability scores
Unresolved
No long-term or workplace follow-up
Skill retention and real project outcomes remain unknown
Watch the background, then check the evidence
Illustrative photo · not a video frame
UNESCO · 11 min 27 sec · Background explainer
UNESCO Talks: Interview with 2024 Nobel laureate Professor Daron Acemoglu
YouTube loads only when you choose; playback starts when you press play.
Economist Daron Acemoglu discusses how choices about AI affect workers, inequality and society. Pair the interview with our comparison of task exposure, hiring and skills evidence.
Evidence check: An expert interview from 2024, offering a perspective on policy and work rather than a measured forecast of job losses.
Published 2024-10-14. Source checked 14 October 2026. Check captions on the publisher’s video page.
Read the AI and jobs guidePhoto: Dylan Gillis / Unsplash · Licence
Continue the story
Read next
Continue exploring knowledge work: compare the finding and its evidence limits.
A slower start helped people write better with AI—and the benefit carried into the next task
A Microsoft Research preprint randomised 297 US workers to an AI writing interface with or without planning prompts. Quality gains were small, and the study did not test long-term workplace performance.
Read the next reportExplore the full source and evidence record
What this means for people
- For managers, teachers and professional teams, the study supports testing workflows that require a human problem frame, alternatives and a final evidence check before accepting fluent AI output—not banning AI or assuming every use weakens thinking.
Global context
Participants were recruited through a Chinese online research platform and completed standardized proposal tasks. Employment norms, language, AI products and organisational accountability vary internationally, so workplaces elsewhere should test structured workflows locally rather than treating the effect size as universal.
What the evidence does not yet show
- Four separate experiments did not estimate the full proposed pathway in one sample
- Online Credamo samples and short simulated innovation tasks
- Compound workflow manipulation changed response detail, effort instructions and responsibility
- Metacognitive disengagement and offloading were self-reported
- Expert ratings were reliable but partly subjective
- No workplace productivity, delivered-project, longitudinal or skill-retention outcomes
- AI literacy was measured, not experimentally trained
What to watch next
- Field experiments in real teams and longer projects
- Workflows that hold response detail constant while varying delegation
- Objective implementation and downstream quality outcomes
- Repeated-exposure and later skill-retention tests
- Randomized evaluation of critical AI-literacy training
- Replication across occupations, languages and AI systems
Living evidence record
Impact record IAI-1I9T0F3
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
28 October 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what Frontiers in Psychology published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 28 October 2026
Research question, topics and regions
How high-assistance and high-delegation AI workflows affect task-level planning, monitoring, reflection and expert-rated proposal quality
Topics: Knowledge work · Human-AI collaboration · AI literacy · Metacognition · Creative work · Workplace skills
Regions: China · Asia
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.