The paper looks for an exercise environment where the teacher program would enable to fix parameters of random generation so that the generated tasks have desired pedagogical properties. In particular, it should be possible to prescribe qualitative content and size of the solution and its parts, opportunities to make certain errors, probabilities of different decisions to be made by the student. We analyse two existing environments for algebraic exercises in Propositional Logic and identify what additional options are necessary and what computational engines or precomputed resources could produce the desired properties.