StewAI Blog · Research · 14 min read
Humans Draw the Lines: What 106 Experiments Reveal About Human-AI Teams
The largest meta-analysis on human-AI collaboration found the average team performs worse than the best of either alone. The fix is not better prompting. It is structure. We walk through the academic evidence and map it to workflow design.
Frequently asked questions
Do human-AI teams outperform humans or AI alone?
On average, no. The largest meta-analysis of 106 experiments found the average human-AI team performed worse than the better of human-alone or AI-alone. The teams that won assigned roles deliberately before the work started.
What is the jagged frontier of AI ability?
AI is unexpectedly strong at some tasks and unexpectedly weak at others, with no smooth boundary. Because the frontier is jagged, who does what has to be decided per task rather than assumed.
How do you make human-AI collaboration actually work?
Decide the division of labor up front and encode it as structure: define which steps the model handles, which a human handles, and where the checkpoints sit, instead of relying on better prompting in the moment.
Read the full article on StewAI