Why AI stays narrow

The question

Statens Serum Institut gave all staff a company GPT tool. Use stayed narrow: some people used it for a few tasks, many not at all. The usual explanations blame people: resistance, missing training, weak management. I asked whether the work itself explains it better.

What I did

A qualitative single case study: 13 semi-structured interviews with researchers and support staff across very different roles, analysed with template analysis into 35 codes, 9 themes and 3 dimensions. Four interviewees did two structurally different kinds of work, which gave a within-person comparison: the same person, different tasks, different adoption.

I then built the tool the interviewees described, a scope-specified reference assistant in n8n, and compared its behaviour with the generic tool.

The theory behind it

The thesis combines organisational routine theory with the dynamic capabilities literature. Routine theory says work is carried by recurring patterns of action, and a new tool has to fit into them to stick. Dynamic capabilities explains why an organisation's ability to change those patterns is built and learned over time, and cannot simply be bought.

Put together, they predict what the interviews showed: a generic AI tool is absorbed where it fits the routine and where its output can be verified cheaply, and nowhere else. The method was an abductive single case study with template analysis, and a design science prototype to test the remedy the users described.

What I found

1. Adoption depends on the task more than on the person. Four interviewees used AI in one part of their job and not in another, so the same person gives both outcomes.
2. AI lands where its output is cheap to check. Across roles, use converged on finding things, polishing text and writing code.
3. Not using it is four different problems, each with its own fix, and nobody in the organisation was turning individual learning into shared practice.
4. The users already describe the fix: AI scoped to one job, anchored in their own material, built into the tools they use, with clear rules on what it is for.

What it means for a company

More training and encouragement will not move the number. Pick routines where results are easy to verify, build a scoped tool for each, and give a small cross-functional team the mandate to find the next one. Implement it routine by routine.