To study the long-term usefulness of genetically-modified agriculture via herbicide-tolerant crops, a simulation model is built by focusing on the fundamental environmental feedback mechanisms. The most critical mechanism is the evolution of resistance in weeds via natural selection. Agricultural sustainability is investigated under different policies and scenarios, in comparison with conventional crops under two herbicide strategies.  In the first strategy, herbicide amount is a function of weed density; in the second it is constant. It is found that superweed emergence increases the rate of resistance evolution in weeds. Under the constant herbicide strategy, GM crop is more effective than the conventional crop. However, this strategy results in a higher rate of resistance development and more herbicide usage than the first strategy. In terms of long term cumulative yield losses, rate of resistance development and herbicide usage, the best policy is discovered to be planting conventional crops under variable herbicide strategy.