A method for discovering optimal algorithms is provided. The method comprises defining a starting condition for a program defining a number of program targets, wherein the program targets map program user-specified inputs to expected outputs, defining a number of program objectives, and defining a number of program constraints. An iterative optimization problem is then initialized to solve for the resulting program. A determination is made if defined termination conditions have been met according to the program targets and optimization objectives. Responsive to the defined termination conditions not being met an update to the program is selected by addition or subtraction of operations, the updated program is evaluated according to the program targets, the number of program objectives are evaluated, and the optimization problem is re-iterated. Responsive to the defined termination conditions being met, the program is reassembled into an instruction set.