Commercial autonomous agents such as Claude and Codex now run for hours or even days to complete tasks, and along the way they show complex behavior: they plan, reason, use tools, recover from errors, coordinate with subagents, and communicate with users. We use the word behavior, as in the study of human behavior, for the full range of what an agent does during runtime. This behavior spans three levels: what agents do and how they do it, what humans do in response, and how the two work together through instructions and corrections. All three generate vast behavioral data such as execution logs and interaction traces. Yet existing approaches read this data largely for outcomes: benchmarks tell us whether an agent succeeds or fails, but not what it did or how it did it.
Understanding what and how is what people actually need. It lets agent developers and model trainers debug failures, compare architectures, and filter training data; it lets agent users and deployment engineers watch production agents to understand safety, cost, and reliability risks. For agentic models, the trajectory is both the training data and what the reward scores. Interpreting it therefore sits inside the training loop, deciding which rollouts are safe to reinforce and flagging reward that reflects a verifier exploit rather than real skill. But the field still lacks the vocabulary, methods, and tools to describe and analyze agent behavior at scale. Humans cannot read through thousands of log entries; they need patterns, summaries, and explanations, in other words interpretation, and we do not yet know how to scale it.
IAB works toward an interpretive science of agent behavior. It treats behavior across these three levels as the object of study and proceeds in two steps: first gathering the community to identify the problem space and emerging challenges, then bringing the broad set of methods that social scientists have developed, such as grounded theory, qualitative analysis, error analysis, corpus analysis, trace analysis, and red-teaming, to read meaning from this data, discover categories from it, and count them. IAB bridges two communities: social science and HCI contribute the interpretation methods, while AI contributes the problem space of evaluation, governance, alignment, and responsible AI.