For those on the fence:
1. What's the specific fear? (data leakage, prompt injection, rogue actions?)
2. What would it take to trust an agent with your real accounts?
3. Are you running agents on burner accounts / sandboxed data instead?
Building something in this space and want to understand the actual blockers.