Allocating Thought: Information Availability and Articulability in Enterprise AI

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

A common assumption in human-AI collaboration is that better outcomes require more context. In knowledge work, however, some of the information needed to complete a task is difficult to access or express. In this paper, we introduce a typology developed through a Research-through-Design study in corporate goal-setting that distinguishes five layers of contextual information based on (a) system accessibility and (b) user articulability. The typology helps identify when a use case depends on information the system cannot access, and what that implies for design. We argue that designing effective human-AI collaboration requires allocating cognitive work appropriately between the user and the system, deciding where cognitive work can lead to meaningful insights. When crucial context is unavailable to the system, cognitive work must shift toward the user, letting them think with the system rather than the system for them. Based on this perspective, we propose a lightweight diagnostic for deciding when AI systems should generate outputs, elicit inputs, or scaffold user understanding. The paper provides practical guidance for designing enterprise AI systems that align with users’ cognitive processes in real-world settings.