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

arXiv: https://arxiv.org/abs/2603.26942 · HTML: https://mlsanity.com/paper/2603.26942/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal and embodied agents, one of 16 directions in LLM agents and tool use.
- Builds on 4 of its 9 references that are in the map.
- First posted on arXiv 27 Mar 2026.

## Placed in

- LLM agents and tool use / [Multimodal and embodied agents](https://mlsanity.com/agents/multimodal-and-embodied-agents/)
- LLM alignment / RLHF / preference optimisation / [Process reward models for reasoning](https://mlsanity.com/alignment/process-reward-models-for-reasoning/)

## Builds on (most cited first, 4 of 4)

- [Reflexion: Language Agents with Verbal Reinforcement Learning](https://mlsanity.com/paper/2303.11366/) (2023)
- [AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation](https://mlsanity.com/paper/2308.08155/) (2023)
- [ChatDev: Communicative Agents for Software Development](https://mlsanity.com/paper/2307.07924/) (2023)
- [When Your AIs Deceive You: Challenges of Partial Observability in Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2402.17747/) (2024)

## Built on by (most cited first, 3 of 3)

- [Autonomous Research Agents: A Survey of AI Scientists and the Verification Gap](https://mlsanity.com/paper/2608.05179/) (2026)
- [Phantom Guardrails: When Self-Improving Agent Harnesses Fix Failures That Never Happened](https://mlsanity.com/paper/2607.13083/) (2026)
- [What Coding Agents Leave Unsaid: Measuring Task-State Gaps Between Execution and Reports](https://mlsanity.com/paper/2610.20003/) (2026)

## Abstract (arXiv)

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.