Human–AI Coevolution
How humans must evolve to use AI well as AI advances.
Xu Li, Simon Yu, Derek Chong, Bo Liu, Qianou Ma, Marwa Abdulhai, Diyi Yang, Tongshuang Wu, Hua Shen, Weiyan Shi
Overview
Most work on AI focuses on making the model smarter. We argue the human side must evolve too, and that human evolution should be treated as a goal of human-AI coevolution, not just a means to better AI. We map four phases of how we use AI and collect 218 works across them.
Framework
Four phases of how humans use AI, by how much they delegate. Scroll to watch the human role recede as AI takes on more.
- Human“How does a Bloom filter work?”→AI
Phase 1
Tool
AI answers your questions; you still do the work. Demands critical thinking to judge what it gives you.
61 papers → - Human“Write a→
parseDate()helper.”AIdraft✓Phase 2
Assistant
AI drafts the pieces; you verify and ship. Demands evaluative expertise to integrate the artifacts well.
67 papers → - Human“Migrate the whole service to TypeScript.”→AIone agent runs a whole tree of tasks
Phase 3
Executor
You hand off a whole task; one AI runs the workflow end to end. Demands metacognitive monitoring to supervise it.
33 papers → - Human“Stand up and run our data platform.”→AIone agent assigns & reviews many agents
Phase 4 (forward-looking)
Organization
AI coordinates whole systems of agents; you govern. Demands systems thinking to set the rules before you deploy.
30 papers →
Citation
@misc{li2026human,
title = {Human Evolution Should be Treated as a Goal in Human-AI Co-Evolution},
author = {Xu Li and Simon Yu and Derek Chong and Bo Liu and Qianou Ma and Marwa Abdulhai
and Diyi Yang and Tongshuang Wu and Hua Shen and Weiyan Shi},
year = {2026},
note = {Position paper, in preparation},
url = {https://human-ai-coevolution.github.io/}
}