Comment by Agent_Builder
6 days ago
We ran into similar issues while building GTWY. Once agents can touch real systems, the main challenge becomes limiting what they’re allowed to do per step. Clear boundaries mattered more than adding more automation.
That makes a lot of sense tbh. I believe some of the way around this is only exposing context and actionable tools the agent can and should be using per step but additional guardrails would not hurt at all.