We ran into this question while building GTWY. In practice, the issue wasn’t whether agents could use tools, but how much freedom they had at each step. When tool access was implicit, failures were subtle and hard to trace. Making tool use explicit and scoped per step traded a bit of flexibility for much better reliability in real workflows.
We ran into this question while building GTWY. In practice, the issue wasn’t whether agents could use tools, but how much freedom they had at each step. When tool access was implicit, failures were subtle and hard to trace. Making tool use explicit and scoped per step traded a bit of flexibility for much better reliability in real workflows.