Design a modern optical system requires the integration of expertise in widely different fields such as, lens design, thermal and mechanical analysis, sensors, etc. Each one of these introduces a large number of loosely constrained parameters which have to be properly chosen for the optimization of system design. In practice, an exhaustive search of all the possible alternatives is an operational impossibility and various 'short-cuts' to narrow the search are used. Commonly, experts in different fields are called upon to pool their resources to 'truncate-the-search-tree' and arrive at an optimum design. This traditional method of designing optical systems requires a significant commitment of resources and represents major bottleneck in implementing advanced optical systems. We have studied this problem in the context of AI with the hope of expediting this design process. It should, in principle, be possible to put both the optical heuristics (an expert's knowledge and experience) and the deep knowledge (fundamentals of optical system design, NASTRAN, CODE V, etc.) in a knowledge base from which rapid inferences can be drawn. To fully exploit a knowledge base one needs to create a suitable environment. A possible architecture for such an environment for optical system design is discussed here.