From Values to Frameworks: A Qualitative Study of Ethical Reasoning in Agentic AI Practitioners

Practitioners use three distinct ethical reasoning frameworks rather than a single set of values: AI practitioners designing autonomous systems justify their decisions through either a Customer-Centric approach (focused on business and legality), a Design-Centric approach (emphasizing technical safeguards), or an Ethics-Centric approach (prioritizing broader social good). This means that ethical reasoning in AI development is more complex than simply asking "what are the right values?"—different practitioners may reach different conclusions based on which framework they prioritize.

Business interests and technical solutions alone are insufficient for responsible AI: The study reveals that practitioners often rely on customer-centric reasoning (business interests and legality) or design-centric reasoning (technical fixes), but the authors argue these frameworks must be actively balanced with ethics-centric considerations that look beyond compliance to broader moral responsibility. This suggests companies need deliberate processes to ensure ethical concerns aren't overshadowed by commercial or technical considerations.

Autonomous AI systems require intentional management of diverse ethical perspectives: Rather than expecting individual practitioners to naturally balance competing ethical frameworks, the paper recommends that AI providers must "actively manage how these diverse reasoning frameworks are represented in their decision-making processes." This implies that responsible AI governance requires structural changes—like diverse ethics review boards or formal processes that ensure multiple perspectives are considered—not just hiring ethically-minded people.

Current approaches to responsible AI may miss how practitioners actually reason: Prior research has focused on what values people say they prioritize, but this study shows practitioners actually reason through frameworks that combine multiple considerations. This gap suggests that responsible AI policies based solely on stated values may not effectively guide real-world decision-making in autonomous systems development.