we try not to have too many opinions in how agents should work every model is intensely trained to use certain tools and workflows so we try to match that environment perfectly you can think you came up with something better but it will just underperform