Which two approaches are MOST effective for improving agent trustworthiness in end-user workflows?

Improving agent trustworthiness hinges on transparency and verifiability. Source-backed answers (A) allow users to verify the information's origin, grounding the agent's claims in reality. This is a core principle of Retrieval-Augmented Generation (RAG) systems. Transparent reasoning summaries or action traces (B) provide a window into the agent's decision-making process, a key aspect of Explainable AI (XAI). By showing how it arrived at an answer, the agent moves from being a "black box" to a understandable partner, which is crucial for building user confidence and facilitating debugging in end-user workflows.

C. Hidden tool execution with no explanation is the opposite of transparency and actively erodes trust by obscuring the agent's actions and data handling.

D. Random refusal behavior makes the agent unpredictable and unreliable, which are qualities that are fundamentally detrimental to establishing user trust.

E. Omitting confidence indicators or review paths prevents users from assessing the agent's certainty and leaves them with no recourse for errors, reducing trust.

1. NVIDIA NeMo Framework Documentation. "Retrieval-Augmented Generation (RAG)." The documentation on RAG pipelines emphasizes connecting LLMs to enterprise knowledge bases to provide verifiable, source-backed answers, which directly enhances trustworthiness. (Accessed via NVIDIA Developer Zone documentation for NeMo).

2. Gunning, D., et al. (2019). "XAI—Explainable artificial intelligence." Science Robotics, 4(37). This foundational paper discusses the importance of explainability, including showing reasoning traces, for building user trust in AI systems. DOI: 10.1126/scirobotics.aay7120.

3. Stanford University, CS221 Course Notes. "Responsible AI: Fairness, Explanations, and Privacy." Lecture notes often cover how transparency in reasoning and data sourcing are critical components for building trustworthy and responsible AI systems.