The Saturation Trap and the Subjectivity of Intervention Timing: Why Affect-Based Triggers and LLM Judges Fail to Time Interventions on Autonomous Agents
2026-06-02 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors studied how to decide the best moment to stop or interrupt AI agents when they struggle during tasks. They tested different methods using a complex emotional-simulation tool and compared their results to human judgments. They found that AI signals often stayed stuck at maximum frustration, making simple threshold methods trigger too often. They also discovered that small language models rarely detected problems, and even advanced models needed lots of context but were still not very accurate. Most importantly, humans often disagreed on when to intervene, showing that timing interruptions is difficult and unreliable to pin down precisely.
Autonomous AI agentsRuntime safetyIntervention timingAffective dynamicsHEART engineSWE-bench-VerifiedLLM judgesInter-rater reliabilityKrippendorff's alphaCohen's kappa
Authors
Manvendra Modgil
Abstract
As autonomous AI agents move from conversational systems to long-horizon software execution, runtime safety layers that decide when to interrupt an agent have become essential. We study this timing problem using a continuous 18-dimensional affective-dynamics engine (HEART) as a diagnostic probe, evaluating four intervention trigger families - absolute state thresholds, composite state-action patterns, regex reasoning-feature extraction, and zero-shot LLM-as-judge - against human-annotated intervention points on SWE-bench-Verified debugging traces. We report three findings. First, a State Saturation Trap: agents show no recovery signal under sustained difficulty, so modeled frustration quickly crosses the threshold and stays at its maximum, converting threshold-on-state triggers from moment detectors into near-constant indicators that fire on 39-83% of actions across five trajectories. Second, a capability-and-context floor for LLM judges: a small model (gpt-5.4-mini) never fires, while frontier and cross-vendor models escape the zero-firing floor only with full-trajectory context, and even then reach only F1 0.17-0.40 at up to 90x the cost. Third, and most importantly, the supervised target is not reproducible among humans: three trained annotators using one rubric on a 56-action trajectory agree on where to intervene only slightly above chance (location Krippendorff's alpha = +0.047; best pairwise Cohen's kappa = +0.349) and not at all on intervention type (pause degenerate; clarify below chance; reflect only alpha = +0.226). We conclude that intervention timing is a low-reliability construct, making single-annotator F1 an unsuitable optimization target. Our contribution is the joint mapping of this problem across human inter-rater reliability, four detector architectures, a cross-model LLM-judge sweep, and a reproduced saturation effect, rather than any single detector's accuracy.