Longitudinal convergence of practitioner trust and measured quality

Determine whether practitioner trust in agent-generated architecture documentation converges on measured artifact quality under sustained use.

Background

The Agentic Software Architecture Workflow (A-SAW) produces architecture documents whose measured structural quality is high, but the evaluation found that practitioners’ trust remains conditional. Trust depends on factors not captured by the rubric, including the abstraction level at which the agent operates, the amount of inference required beyond repository data, and the accumulation of reliable generations over time.

Because the study used a single-session evaluation, it could not assess whether repeated exposure to reliable agent-generated documents causes practitioner trust to converge toward the documents’ measured quality. Resolving this issue requires longitudinal observation of trust, review behavior, and artifact quality during sustained use of A-SAW.