Settings an admin sets once for every rep: model, guardrails and shared rules.
Settings → "AI Training".
Sections tagged "Organization-level" apply to everyone in the org and only admins can change them.
The model choice is one of them.
Guardrails: daily send ceiling, never double-tap an account, hold on sensitive triggers, aggressive warm-up.
Org settings holds the rest: visibility, collision policy and account scoping that agents respect.