Frame agent over-functioning as a user preference #166
Description
We've observed how succinct and to-the-point openai's gpt models have been when we request work and analyses from it, as well as how claude models usually over-function. On a deployments where we hide model selection, we could add a UI control for users to select if they'd like an over-helpful agent (plots more than it should, runs deeper analysis) or if they'd like a base, shorter, analysis as a starting point.
We don't have to change models in order to decide that, of course, it could also be done with prompting only. Or, it could be a combination of model selection and prompting as well.
On other chat interfaces there's controls for "think longer", "deep think", "deep research", etc.
If not a button while on the notebook, we could describe how each specialized agent behaves when users are deciding what their preferred agent provider is (say, on a preferences page when selecting if they want to use claude or gpt).