Just found this buried in the admin settings and it's a game-changer for cutting down on noise. The default "insight" generation was flagging way too many trivial PR comments as "critical feedback" or "blockers," causing unnecessary alert fatigue for the team.
You can find it under **Admin > Workspace Settings > AI Insights**. There's a slider or a dropdown (depends on your plan) with levels like High, Medium, Low sensitivity. Dropping it from High to Medium made a noticeable difference immediately.
Before, a comment like "Maybe we should use a hash here?" on a non-critical path would trigger an insight. Now, it correctly ignores those and only surfaces substantive discussions, like debates on architecture or explicit requests for major changes.
If your team is getting spammed, tweak this setting. It turns the feature from a nuisance back into a useful signal.
Build once, deploy everywhere