In power systems with high penetration of power electronics, grid-forming control is proposed to replace traditional Grid-Following Converter (GFL) in order to improve the overall system strength and resist small-signal instability in weak grids by directly forming the terminal voltage. However, sufficient headroom of both active and reactive power must be made available for Grid-Forming Converter (GFM) to operate, potentially leading to sub-optimal operation in steady states. This presents a new research problem to optimally allocate between GFM and GFL to balance the ability of GFMs to improve the grid strength and the potential economic loss resulting from reserved headroom. An optimization framework under software-defined grids is proposed, for the first time, to dynamically determine the optimal allocation of GFMs and GFLs in power systems at each time step of system scheduling according to system conditions, which ensures both system stability and minimum operational cost. To achieve this, the system scheduling model is expanded to simultaneously consider the constraints related to active and reactive power reserves for GFMs, as well as the system level stability. Case studies conducted on the modified IEEE 30-bus system demonstrate significant economic benefits in that the optimal proportion of GFMs in the power system can be dynamically determined while ensuring power reserve and grid stability constraints.