Despite the potential of augmented reality (AR) to support and guide complex industrial tasks, the technology is still not broadly applied. One possible reason for this is that the creation of AR content is highly complex and requires programming skills and deep spatial knowledge. AR authoring tools can help address this complexity by enabling non-developers to create self-sufficient AR content. Therefore, this paper proposes a theory-driven design for AR authoring tools that allows non-developers to create self-sufficient AR-based instructions to support complex tasks. Based on ten interviews with experts working with AR authoring tools and a following focus group with eight participants, we propose three design principles for future AR authoring tools in the engineering context. These design principles are instantiated in two prototypes of different richness and evaluated in an experiment with 23 students. Our study shows that the cognitive load is slightly increased when using the extensive AR authoring tool, but it also shows that significantly better results can be achieved with the extensive AR authoring tool. We contribute by providing design principles for AR authoring tools for creating AR-based instructions, which extend the existing AR authoring research in the industrial context.