customer support operations

Abstract: Large language model agents are often deployed in environments where new knowledge, tools, and workflows keep arriving. Such an agent is therefore expected to adapt continually, learning new tasks as they arrive without forgetting earlier ones. This resembles the continual learning problem, but at a pace that conventional gradient updates cannot sustain. Nonparametric self-evolving agents are a natural alternative: they adapt by rewriting their own harness, such as prompts, memory, and skills, from task feedback while the model weights stay frozen. We study whether such agents are effective continual learners across four drifting task streams that shift in topic, workflow, or domain (built from BIRD, BFCL, and ScienceWorld). Evaluating three representative methods spanning experience retrieval and self-modification (ReasoningBank, GEPA, and Meta-Harness), we find that although self-evolution improves aggregate performance on most streams, higher plasticity consistently comes at the cost of stability. Specifically, self-modifying methods apply global edits to a shared prompt or workspace, interfering with unseen tasks and overwriting earlier knowledge as the stream shifts. Conversely, similarity-based memory retrieval localizes updates across distant domains, but still retrieves foreign memories on unseen tasks and suffers from retrieval collision when similar queries require conflicting workflows. These findings show that current self-evolving agents remain incomplete continual learners.

Abstract: Large Language Model (LLM) agents are increasingly deployed in high-stakes settings such as industrial maintenance and equipment fault troubleshooting, where workers occupy a variety of roles. A capable agent must therefore act in a way that is calibrated to user's role: taking actions and providing information that respect the role's knowledge and capability boundaries. Unlike coding, where mistakes are usually recoverable, agent responses in these settings are enacted on physical equipment, and can therefore cause irreversible equipment damage, production loss, or personnel harm. Existing benchmarks, however, largely overlook the need for agents to infer what a role intends and acting only through tools that role may legitimately use, a capability which we term Perspective Awareness. To this end, we introduce ReFract, a benchmark of 150 expert-validated entries in which an agent must act differently in response to the same query depending on user's role. Entries of ReFract are grounded in anonymized queries from domain support conversations, against which we construct Text World Models that simulate the agent's operating environments and assemble perspective-aware action trajectories. State-of-the-art LLMs solve at most 69% of the tasks with more than 50% of their trajectories contain attempts of taking perspective-violating actions. ReFract exposes perspective awareness as a distinct, largely unsolved axis of agent evaluation and motivates agents that calibrate not just how to act, but for whom.

Abstract: Detecting a silently struggling kiosk user is only the first step; deciding whether, when, and how to help depends on scenario knowledge usually buried in model weights and thresholds. We propose a scenario-knowledge-driven pipeline: a single scenario knowledge document, human-authored and version-controlled, configures sensing, constrains LLM reasoning, and shapes a graded intervention proposal. Narration, assistance-need assessment, and proposal are kept separate for independent audit. As proof of concept, we replay two recorded kiosk sessions offline, chosen before the runs for their struggle evidence and retrospective detail. Both cases support what the design promises: checkable reporting and measured escalation. Across 95 updates, every sentence of the append-only narration cites the primitive events underlying it, and the rule layer detects 12 of 13 and 7 of 7 annotated struggle episodes under a strict criterion. The assessor de-escalates on recovery and reaches the top rung exactly once, under maximally converging evidence. At the decisive help-seeking turn, narration, assessment, and the participants' retrospective accounts converge. The appropriateness of these interventions, the pipeline's restraint on sessions without struggle, and the document's transfer to a new scenario frame the agenda.