Despite significant advancements of Artificial Intelligence (AI) technology, its effective integration in the interpreting field has progressed at a slower pace. Existing products still have room for improvement in efficient application of technologies and better aligning with interpreters’ needs. This study identifies three key challenges in Computer-aided Interpretation (CAI) tool design: (1) Interference, including the distraction caused by CAI tools as an additional visual input and the impact on interpreters’ Ear Voice Span (EVS); (2) Overreliance, the risk of error propagation from machine-generated content to interpreters’ output, as well as underperformance when the system fails to meet interpreters’ expectation; and (3) Insufficient support for interpreters’ needs. To explore the solutions to address these limitations, this study discusses best practices for ergonomic CAI tool design, including enhancing controllability for intrusive features, improving interaction ergonomics through non-intrusive design, extending support to pre- and post-process, and incorporating AI usage disclaimers and built-in user feedback mechanisms. For each practice, this study presents proposed designs and elaborates their implementation. As a practical example, this paper proposes the Help Button Prototype, a targeted real-time lookup mechanism that allows interpreters to request terminology and research support on the fly, leveraging backstage transcripts that prioritize accuracy over speed. By shifting from machine-centered automation to interpreter-driven augmentation, CAI tools can function more as collaborative partners rather than intrusive distractions, which underscores the importance of design philosophy that moves beyond the approach of “humans-in-the-loop” to “humans-at-the-heart”.