Questions about the research

Workability asks whether a task can be carried through in a form that is sufficiently usable, understandable and accountable for its setting. It is different from what a GenAI model can do, and it does not mean full automation or that an arrangement would work elsewhere. In the settings studied, workability depended on how tools, information, the task, organisational conditions and human responsibility were connected, not on access to a capable tool alone. Thesis Sections 1.8 and 7.6

The Staff Development Fund provided an allocation for each staff member’s professional development. Part of my job was preparing statements showing each person’s spending against that allocation. That reporting work is the workflow studied in depth. Thesis Section 4.2.1

The thesis builds on established research on technology in work, workarounds and invisible work, and does not claim a new general theory or method. What it adds is a bounded way of keeping three questions apart in GenAI-supported university work: when a difficulty changes how a task can be done, when an adaptation becomes part of regular work, and when work behind an output can be called invisible labour. Methodologically, it compares what an account of one workflow and interviews with eleven staff can each establish. Thesis Sections 1.6, 7.7.1 and 7.7.2

Both, and that is part of what the case shows. Python did the repeatable preparation, and Copilot came in later to query files that had already been prepared. Copilot access alone would not have solved the original problem. The records first had to be linked, structured and kept traceable, and some answers still had to be checked and explained afterwards. The case therefore shows GenAI-supported work depending on work before and after the interface. Thesis Sections 4.5, 4.7 and 7.3

No. I supplied selected prepared files to Microsoft 365 Copilot and asked questions about them in plain language. The thesis describes this as Copilot-supported querying over prepared information. Retrieval-augmented generation (Lewis et al., 2020) offers only a limited comparison for supplying selected files through an available interface. Thesis Sections 2.7 and 4.5

In this thesis, workaround formation means a goal-directed alternative route, developed in response to a recognised obstacle, that becomes part of the continuing work needed to complete a task. In the SDF case, a recurring reporting problem led to an arrangement that became part of regular reporting. The workaround was the whole connected arrangement, including my checking, not Python or Copilot on its own. It reorganised part of the reporting work without replacing the wider university process. It is classified by what it did and whether it became part of continuing work, not by where it came from. Thesis Sections 2.5, 2.9.2 and 4.7

I did not measure the arrangement’s net effect on time, cost or workload. The study collected no time-motion data, comparative workload records or cost measures. The SDF case shows work being redistributed: some repeated preparation moved into Python, while checking and explanation remained. The interview accounts were mixed. Some participants described more demanding review, while others reported time savings on particular tasks. These are participants’ reports and are not combined into an overall effect. Thesis Sections 6.7, 7.5 and 7.8

It is one recurring workflow examined in depth. Its value is that it lets the path from a problem to a changed way of working be traced, which the interviews could not show in the same detail. Eleven interviews broaden the comparison. Neither shows how common these experiences are or that the arrangement would work elsewhere. Thesis Sections 3.2, 7.2 and 7.8

They are not a representative sample. Eleven staff volunteered after an invitation in the university’s Microsoft 365 Copilot Teams channel, so the group may lean towards people interested in GenAI. The study makes no claims about prevalence. Thesis Sections 3.3.1 and 3.7.2

I used thematic analysis, starting from a codebook based on the research questions, framework and literature, and revising it in response to the interviews. The codebook was tested on four transcripts, then applied to all eleven in NVivo using eight parent codes. Codes organised the material but were not findings. Each passage was read again in its transcript before use. I did all the coding myself, and memos, a codebook change log, evidence-selection records and claim matrices document the decisions. Thesis Sections 3.3.3, 3.6.2 and 3.7.2

Copilot gave bounded help checking coded-passage boundaries and links between evidence and claims. Its suggestions could influence what drew attention, so they were checked against the transcripts. I applied all coding and made every interpretation. Thesis Sections 3.6.2 and 3.7.2

This reflects what the evidence lets me trace, not how useful the changes were. Some were reported to be in use, such as SP05’s two internal agents and SP06’s attendance workbook, and I treat these as changes in use, not proposals. Others were being tested, explored or proposed. The interviews gave only partial detail about the problem each change addressed and how it altered the earlier work, so they could not show workaround formation in the narrow sense used here. That narrow sense is a choice for this study. Alter’s broader definition also includes temporary workarounds. Thesis Sections 2.5, 6.5 and 7.4

Not everything. Some friction came from other systems. The GenAI-related concerns included tracing answers that sounded authoritative, understanding generated code, checking generated language before sending, and correcting plausible but wrong course-code mappings. Model errors and review work appear in earlier research. The thesis shows what they involved in these tasks. Thesis Sections 5.6, 7.3 and 7.9

Look at the whole work arrangement, not only whether a tool is available or an output looks polished. That means asking who prepares the information, checks the results, signs off wider use, explains results, supports the people using it and maintains the arrangement, and whether that work is accounted for. Checking should be matched to the risk and responsibility involved. These are questions for local assessment, not a tested recipe for other settings. Thesis Sections 7.7.3 and 7.8

No. It found less visible work in all three settings: preparation, checking, correction, explanation and learning behind the outputs. Calling that invisible labour would need evidence that the work was not given enough time, support or recognition, and the evidence did not establish that. Thesis Sections 2.6, 4.7 and 7.5