Constraint Programming Next Challenge: Simplicity of Use Jean-Francois Puget ILOG, 9 avenue de Verdun, 94253 Gentilly, France puget@ilog.fr 1 Introduction Constraint Programming (CP) is a healthy research area in the academic com- munity. The growing number of participants to the CP conference series, as well as the number of workshops around CP is a good evidence of it. Many major conferences have a CP track, both in artificial intelligence, and in operations research. The existence of several commercial companies that offer CP tools and services is a further evidence of the value of CP as an industrial technology. ILOG is one of such companies. One of our uniqueness, as far as CP is concerned, is that the research and development team that produces our CP products is also responsible for the development of our mathematical programming (MP) tool, namely ILOG CPLEX. This provides a unique opportunity to contrast the way these products are developed, marketed and used. In this paper we argue that current CP technology is much too complex to use for the average engineer. Worse, we believe that much of the research occurring in the CP academic community makes this even worse every year. The rest of the paper provides evidence for this claim, and suggests ways to address the issue of simplicity of use by looking at how a similar issue has been addressed in the mathematical programming community. 2 A Comparison Between Math Programming and Constraint Programming A technical comparison between mathematical programming and constraint pro- gramming shows many similarities [8]. If we look at it from the angle of an indus- try user, then the two approaches also look very similar. For instance, problems are modeled in a very similar way, using variables, constraints, and an objective function. There are differences though. For instance, the set of constraint types that can be used is usually restricted in math programming to linear or quadratic constraints. CP systems usually support a much richer set of constraint types. There are also differences in the algorithms used under the hood to solve a prob- lem once it is modeled. However, our point is that a typical industry user is not interested in understanding the differences in the way solutions are computed with a CP system as opposed to the way they are computed with an MP sys- tem. What matters is to know if one technology is applicable to the problem at M. Wallace (Ed.): CP 2004, LNCS 3258, pp. 5–8, 2004. c  Springer-Verlag Berlin Heidelberg 2004