I’m pleased to see this debate is beginning to take shape. These recommendations from JISC are important:
- Clear ownership of the AI budget.
- Real-time visibility into usage and cost.
- Agreed thresholds and usage limits, instated at roll-out and adjusted as required.
- Transparency built into contracts from the outset.
- Adequate controls and user-based roles and limits, so usage can be managed at the individual and department level.
- Reasonable exit terms, so value can’t be eroded by future product changes.
The thing I’m getting monomaniacally focused on is how we make the opportunity costs visible. When we are deciding how much to spend on token use and for what we need to find a way of surfacing the other things that money could be spent on.
