When AI will able to take as input a bunch of legal documents and scripts of interviews with customers, to automatically process them, and to output software that solves the problems posed in the input, adhering to the restrictions and requirements that come of it, then, human software engineers, architects, analysts and developers will not be needed. But I think that this is vary far beyond today's AI's capabilities and I doubt that today's AI techniques (neural nets, machine learning, classification etc) will be proved suitable or enough for this task. The scenario of an AI interacting with users, "reading" legal documents and then training it self by trial and error until it produces the "right" software, does not seem quite realistic to me. But maybe my imagination is short...