Agile software development research increasingly treats artificial intelligence (AI) as a knowledge-creating agent yet rarely examines how its outputs become credible project knowledge.
Grounded in the knowledge-based view (KBV), this paper adopts a two-study design. Study 1 systematically reviews 33 Q1 journal articles, using the Antecedents-Decisions-Outcomes (ADO) framework to map reported patterns and a problematisation analysis to derive six theory-informed, structured factors. Study 2 then applies fuzzy Decision-Making Trial and Evaluation Laboratory (DEMATEL) with 52 screened software practitioners to examine their perceptions of the directional relationships among those factors.
Study 1 establishes four problematisations and conceptually derives six factors shaping knowledge outcomes: AI intensity of use, explainability and transparency, team decision autonomy, trust, governance tightness, and agile performance. Study 2 indicates that trust has the highest prominence in the pre-specified network.
The paper extends the KBV through a problematisation that distinguishes computational outputs from credible project knowledge and derives structured factors shaping knowledge outcomes in AI-embedded agile projects. The paper also develops an agility paradox linking reactive, efficiency-oriented AI adoption to potential constraints on proactive agile capabilities.
AI-generated outputs should be treated as provisional knowledge that requires contextual review. Software firms could pilot a lightweight validation record for consequential AI-supported decisions, review these records during retrospectives, and calibrate governance.
The paper's originality lies in reframing AI in agile software development work as a question of knowledge credibility, decision authority, trust and governance.
