Analyzes 73k Reddit posts via Value Sensitive Design to surface 21 user-prioritized values grouped into six clusters (e.g., Autonomous, Dependable, Affordable Operation) for autonomous AI agents.
Value-Sensitive Delegation is the concept that human values in AI agent use attach primarily to the operating conditions users set around a run—such as cost, reach, and oversight—rather than to the agent’s final outputs. Analysis of 73,093 Reddit posts about OpenClaw reveals that values are usually met when describing agent delivery but mostly unmet when describing supervision, shifting the focus of value alignment from model performance to user-configurable boundaries.
The study identifies 21 values clustered into six groups: Autonomous, Dependable, and Affordable Operation, Bounded Reach, Reviewability, and Equitable Access. The most frequent values were autonomy, dependability, and affordability, with the largest groups accounting for over 70% of posts. Key findings indicate that resource accounting and risk exposure were the least frequently met values, highlighting the importance of designing agents with clear, user-settable limits on spending, access, and approval workflows.
This material presents an empirical, value-centered analysis of how people delegate work to everyday autonomous AI agents, drawing on a large corpus of 73k Reddit posts associated with OpenClaw. Rather than evaluating agents primarily through task-completion metrics, it applies Value Sensitive Design (VSD) to treat user discourse as evidence of normative expectations: what users want agents to do, avoid, justify, or leave under human control. The result is a grounded account of delegation as a social-technical practice, where users are not only consumers of capability but also stakeholders in risk, cost, transparency, and accountability.
Its central contribution is a structured taxonomy of 21 user-prioritized values, organized into six clusters, including Autonomous, Dependable, and Affordable Operation. This moves the discussion from a single, abstract alignment goal to a multi-dimensional value space in which competing expectations coexist. For example, users may want agents to act proactively without constant supervision while also expecting reliable behavior, bounded spending, and clear explanations when actions have consequences. By making these values explicit, the paper provides a diagnostic vocabulary for identifying where agent behavior may feel competent but still misaligned with user expectations.
The work matters because autonomous agents are increasingly deployed in everyday workflows where small failures, unexpected costs, or opaque actions can erode trust quickly. The taxonomy offers a practical bridge between AI systems research, HCI, and responsible design: it can inform agent architectures, evaluation protocols, policy constraints, and user-facing controls that reflect real-world priorities rather than only benchmark performance. In short, it reframes agent alignment as a value-sensitive delegation problem, providing evidence-based requirements for building agents that are not only capable but also acceptable, accountable, and fit for everyday use.