The Fusion Model in higher education
Frontier AI at Research Universities
Frontier AI is arriving at research universities faster than they can absorb it. The below read through leverages the Fusion Model, with evidence from 2025 and 2026, to explore how universities use frontier AI, such as Claude, Gemini and ChatGPT, across teaching, research and operations. Each initiative is placed at one of the five adoption stages, and an activity system at each stage shows the contradiction holding it back.
Explore the findings
Choose a stage to move the lens. The red zigzag marks the contradiction holding that stage back. Then choose a part of the activity system to see what the evidence shows there.
Primary contradiction
Through the activity system lens
In contradictionWhat the findings suggest
- A license is not adoption. Campus-wide access is a Matching decision. Adoption is what happens after it, department by department.
- Clarifying is the bottleneck. Four of the five tensions bite there: guidance, trust and integrity norms lag behind use.
- Routinizing needs recurring money. Only 2% of respondents say new AI costs are met with new money (EDUCAUSE).
- Operations may already be quietly routine. Administrators lead daily use, but a routine innovation stops being reported, so this needs testing.
Five tensions, read as contradictions
In activity theory, contradictions drive change. Each tension in the research is a conflict between two parts of the activity system, and each clusters at a particular stage.
| Tension | Parts in conflict | Stage where it bites |
|---|---|---|
| Better work versus less learning | Object versus Goal: higher-quality work, less learning | Clarifying, in teaching |
| Use outpaces guidance | Tools versus Rules | Clarifying |
| Research integrity and unpublished data | Tools versus Rules and Community norms | Redefining and Clarifying, in research |
| AI as collaborator versus researcher trust | Tools versus Community | Clarifying, blocking Routinizing |
| Dependence and divide | Rules (economic, vendor terms) versus Tools | Matching, and every loop back to it |
About this analysis
Stage placements are interpretations of public announcements, news coverage and surveys from 2025 and 2026, not case studies. A stage belongs to an initiative, not a whole institution: Stanford sits at Matching for campus licenses and at Restructuring for its Biomni research agent at the same time. Evidence skews toward the U.S. and U.K. and toward large institutions.
Sources
- Baytas, C., & Ruediger, D. (2025). Making AI generative for higher education. Ithaka S+R. Link
- Digital Education Council. (2026). AI in higher education global survey 2026. Link
- Kothari, T., Wu, B., Sullivan, J. R., & Goetzel, D. (2026). Building the AI-forward research university. Boston Consulting Group. Link
- Robert, J., & McCormack, M. (2025). 2025 EDUCAUSE AI landscape study. EDUCAUSE. Link
- Tyton Partners. (2026). Time for Class 2026: The AI tipping point. Link
- Bent, D., & Handa, K. (2025). Anthropic education report: How educators use Claude. Link
- Elsevier. (n.d.). Global survey of 3,000 researchers. Link
- Imperial College London. (2025). Google’s AI co-scientist could enhance research. Link
- Stanford Report. (2026). From biology to astrophysics, AI expands the boundaries of discovery. Link
- Stanford University IT. (2026). Stanford AI tools: Latest updates. Link
- The Harvard Crimson. (2026). FAS plans to grant access to Anthropic’s Claude, phase out ChatGPT Edu. Link
- OpenAI. (2026). Accelerating scientific discovery with ChatGPT for Academic Researchers. Link
- Anthropic. (2026). Claude Science, an AI workbench for scientists, is now available. Link
Bring the Fusion Model to your campus
In a hands-on workshop, teams place their own AI initiatives on the five stages and find the contradiction holding each one back.