When adopting AI agents, asking which model is “the best” is not enough. What matters is seeing which model suits which part of the workflow, how much it costs, what quality it delivers and where the human has to review the result.

In this workshop we map the human and agent workflow and tie it to model choice, price and metrics. The aim is to help the team make better decisions: when to use a faster and cheaper model, when a more accurate one, what to measure, and how to optimise without losing quality and accountability.

This is the second view in the enterprise trilogy: once the workflow is visible, you can choose the models, assess the cost and measure whether the agent genuinely helps or merely shifts the work back to the human through review and corrections.

Topics

The model choice map: which part of the workflow needs which model and why
Making price and usage visible: requests, tokens, retries, working time and manual review
Measuring quality: accuracy, usefulness, trust, the number of corrections and the need for human intervention
Optimisation decisions: where you can switch the model, change the prompt, shorten the workflow or add a checkpoint
The human role: where the decision stays with the human and where the agent may help measure, compare or prepare

Outcome

Participants leave with a working map for model choice and cost optimisation: which models to use in which parts of the workflow, which metrics to follow, where the costs arise and which optimisation experiments to run next. If needed, it can also serve as a starting point for bringing in a separate measurement or technical optimisation partner.