Drafting a short task explanation and analysing capability across an organisation are different kinds of AI work. They may need different amounts of context, different response times and different levels of human review.
Thoutex’s AI architecture supports Fast, Balanced and Quality profiles, along with managed AI and provider choice through configuration. The purpose is to make those trade-offs deliberate. A profile name is a starting point for evaluation, not a guarantee about every output.
Begin with the workload
List the AI-assisted activities your team expects to use. These might include drafting learning-map nodes, generating quizzes, suggesting goals or producing capability analysis. Describe what makes an output usable for each activity before choosing a profile.
A short draft that a manager can quickly edit may have different needs from a detailed analysis that must preserve the meaning of several evidence records. Judge the result against the task, including omissions and ambiguity, rather than only checking whether the text reads well.
Compare profiles with the same example
Where the configured workflow allows a profile choice, use the same source material and request for a comparison. Review factual consistency, relevance to the learning level and the amount of correction needed. Also notice how long a person waits before they can continue their work.
Fast may be a useful candidate for small, repeatable drafting tasks; Quality may be worth evaluating for more demanding generation; Balanced provides another trade-off to assess. These are starting hypotheses. A profile that performs well on one activity should not automatically become the default for every other one.
Consider review effort alongside usage
Thoutex captures AI usage and cost information and supports controlled routing of generation work. Use that information alongside your team’s experience of reviewing the results. A response that requires several retries or extensive rewriting has a practical cost beyond its first generation.
Keep a simple evaluation note: the task, the chosen profile, whether the draft was usable and what had to be corrected. Combining usage figures with reviewer observations gives you a fuller basis for choosing a profile.
Keep provider configuration and content review distinct
Managed AI and bring-your-own-key options concern how workloads reach a configured provider. Availability depends on the organisation’s setup; do not assume every provider or profile is enabled in every workspace.
Whichever configuration is used, review learning material and assessments before assigning them. A higher quality profile does not remove the need to check local procedures or the intended answer to a quiz question. Revisit the choice when the workload changes, using examples from the work people actually need to complete.