If you deal with computationally hard problems, I think there's a whole toolbox of utilities with unbelievable performance, in which millions of PhD-holder-hour

If you deal with computationally hard problems, I think there's a whole toolbox of utilities with unbelievable performance, in which millions of PhD-holder-hours have been poured, but that get ignored more often than not. - SAT modeling languages & SAT solvers - same for SMT (satisfiability modulo theories) - Constraint programming - MIP (mixed-integer programming) and its special cases, like - TSP (traveling salesman problem) The latter is maybe more anecdotical (as in: fewer direct applications), but it is emblematic of the whole phenomenon. We see a steady stream of claims that NP-hard problems are "impossible" or take "billions of years to solve". We also see blog posts with the latest ML approach "doing the impossible" and providing reasonable (not optimal) solutions to 50-city TSPs. All this while ignoring that with proper maths, we could solve 50-city TSPs to optimality by hand in the fifties, and now Concorde routinely solves 100k-city instances in seconds on an iPhone.