Cursor IDE
We just upgraded to cursor enterprise and we have team members using different models day in and out.
we would like to know if the tasks the team members are working are solved with model with appropriate intelligence. we don’t want members to solve trivial tasks with 5.6 sol ultra. is there an analysis page where admins can check
task vs appropriate model intelligence is used to know if it we are overusing or under using.
Hi @A_K5 Thank you for the post and congrats on the upgrade! Great question, and this is close to exactly what Conversation Insights was built for.
It lives in the web dashboard under Usage Analytics and is enabled by default on Enterprise plans, so it should already be collecting for your team unless an admin has turned on Disable Conversation Insights in team settings. You can get to it here: cursor.com/dashboard/conversation-insights (admins see the whole team, members see their own data). Full docs: Usage Analytics: Conversation Insights.
Cursor classifies each agent session on device, so no PII or code leaves the machine, across five dimensions:
Write Code (writing code, building features), Ask (understanding the codebase, asking questions), Plan (planning an approach before implementation), Task Automation (delegating git commit, lint, build, create PR).Trivial (single file, minimal change), Low (1-2 files), Medium (3-10 related files), High (cross-cutting concerns, many searches and edits, architectural changes).Low (vague, no code references or constraints), Medium (one of code references, acceptance criteria, or constraints), High (multiple evidence types, well specified).Bug Fix, Feature, Refactoring, Testing, Documentation, Configuration, Ops and Deployment, Styling, Explanation, Performance.Bug, KTLO (maintenance, tech debt, dependency updates), Feature.Task Complexity is the dimension you may be most interested in, and Compare breaks it down by team, group, or individual developer. Reading that next to model usage on the main Analytics page (Messages Sent filtered by model) tells you whether someone doing mostly Trivial and Low complexity work is reaching for a frontier model. These are 2 separate views, but combined, they give you exactly the answers that you’re looking for. Both sides are in the Analytics API if you want to join them yourself: /analytics/team/conversation-insights and /analytics/by-user/models.
Let me know if you have any follow-up questions!
Thanks for the answer @kevinn !
we are looking for something like this, int he conversation insights , it would be better to have something like “what other models could have been used for these type of usage to save cost here” suggestion.
it can also help us monitor what models are actually required for us to enable and disable.
