The Observability Gap: Why Output-Level Human Feedback Fails for LLM Coding Agents

Wang, Yinghao and Wang, Cheng (2026) The Observability Gap: Why Output-Level Human Feedback Fails for LLM Coding Agents. In: CHI 2026 Workshop on Human-Agent Collaboration. UNSPECIFIED. (In Press)

Full text not available from this repository. (Request a copy)

Abstract

Large language model (LLM) multi-agent coding systems typically fix agent capabilities at design time. We study an alternative setting, earned autonomy, in which a coding agent starts with zero pre-defined functions and incrementally builds a reusable function library through lightweight human feedback on visual output alone. We evaluate this setup in a Blender-based 3D scene generation task requiring both spatial reasoning and programmatic geometric control. Although the agent rediscovered core utility functions comparable to a human reference implementation, it achieved 0% full-scene success under output-only feedback across multiple instruction granularities, where success required satisfying object completeness, ground contact, collision avoidance, and scale plausibility simultaneously. Our analysis identifies a structural observability gap: bugs originate in code logic and execution state, while human evaluation occurs only at the output layer, and the many-to-one mapping from internal states to visible outcomes prevents symptom-level feedback from reliably identifying root causes. This mismatch leads to persistent failure mode oscillation rather than convergence. A diagnostic intervention that injected minimal code-level knowledge restored convergence, strongly supporting the interpretation that the main bottleneck lies in feedback observability rather than programming competence. We formalize this phenomenon as a feedback paradox in domains with deep causal chains between internal code logic and perceptual outcomes, and argue that effective human–agent collaboration in such settings requires intermediate observability beyond output-only evaluation. Code is publicly available at: https://github.com/JasperWANG911/CHI_evolve_agent.

Item Type:	Book Section
Faculty \ School:	Faculty of Science > School of Computing Sciences
Related URLs:	https://arxiv.org/abs/2603.26942
Depositing User:	LivePure Connector
Date Deposited:	09 Apr 2026 16:30
Last Modified:	09 Apr 2026 16:30
URI:	https://ueaeprints.uea.ac.uk/id/eprint/102744
DOI:

Actions (login required)

View Item