Human-AI co-creation modes

Human-AI co-creation modes describe how people and GenAI jointly steer a workflow over time, not merely whether a human and AI are both involved. The useful unit is the pattern of agency across the workflow: who decides what work needs to be done, who decides how it gets done, and whether the human remains cognitively engaged enough to build or preserve expertise.

2026-dellacqua-jagged-technological-frontier gave the wiki an earlier behavioral vocabulary for consultant collaboration patterns: Centaurs divide and delegate work between themselves and AI, while Cyborgs tightly integrate AI through the task flow. The key lesson was frontier calibration: close integration can help inside the frontier but amplify wrong output outside it.

2025-randazzo-cyborgs-centaurs-self-automators turns that vocabulary into a workflow and skilling typology. In a field experiment with 244 global management consultants using GPT-4 on a strategic problem-solving task, the authors identify Fused Co-Creation/Cyborgs, Directed Co-Creation/Centaurs, and Abdicated Co-Creation/Self-Automators. The modes differ by who selects what needs to be done and who identifies how it gets done, which makes co-creation a work-architecture construct rather than an attitude toward AI.

The skilling result is the main reason this topic deserves its own page. Fused/Cyborg users built GenAI-related expertise while maintaining task expertise, Directed/Centaur users built task/domain expertise, and Abdicated/Self-Automator users built neither. That means co-creation modes are not just interaction styles; they are learning environments that shape whether AI-mediated work becomes newskilling, upskilling, or bypass.

Connections

• Connects to human-ai-collaboration because co-creation modes are a more precise substructure inside the broad collaboration category.
• Connects to work-redesign because co-creation patterns allocate problem framing, method selection, analysis, iteration, evaluation, and reporting across human and AI.
• Connects to ai-mediated-learning because different interaction modes expose or bypass the practice and explanation work that builds durable competence.
• Connects to ai-induced-skill-erosion because abdicated co-creation can preserve polished output while weakening the human contribution the system may still depend on.
• Connects to human-ai-task-taxonomy because mode claims need explicit task dimensions: task definition, output definition, audit requirement, decision authority, AI contribution, and human persona.

Contradictions & open questions

• 2026-dellacqua-jagged-technological-frontier warned that intensive collaboration can harm performance outside the AI frontier, while 2025-randazzo-cyborgs-centaurs-self-automators finds that Fused/Cyborg work can build AI-related expertise. The design question is when deep integration teaches calibration versus when it accelerates confident error.
• 2025-randazzo-cyborgs-centaurs-self-automators shows that Directed/Centaur users had the highest accuracy, but Fused/Cyborg users built more AI-related expertise. Organizations may need to choose between immediate output accuracy, AI fluency, and domain upskilling rather than assuming one collaboration mode optimizes all three.
• Abdication is not automatically bad. The source explicitly notes that routine tasks, time pressure, or work well inside the model frontier may make delegation efficient; the risk is treating abdication as harmless when the task is also supposed to preserve judgment, domain understanding, or future oversight capacity.
• The current evidence is based on short-term consultant experiments. The wiki needs longitudinal field evidence on whether people switch modes across tasks, whether managers can design for mode choice, and whether co-creation mode predicts durable expertise, productivity, accountability, or worker experience.