Benchmark Radar Day 86: Customer-Service Agents Need Process Checks

Day 86 of Benchmark Radar. The scan fetched 1,430 items, published 755, and marked 389 as recommended.

Author: Koutian Wu; GitHub: ktwu01

Benchmark Radar is live. Open the dashboard, star the repository, or read the paper. The Hugging Face dataset provides the data.

CXBench scores both outcomes and procedures for tool-using customer-service agents across six regulated domains. Rule cards connect obligations to observable trace evidence, including disclosures, consent, and required steps. Reaching the desired business state alone cannot earn full credit. CXBench: A Benchmark for Process-Compliant LLM Agents in Regulated Customer-Experience Workflows.

VibeTraining-Bench evaluates research agents that improve an entire agent system under a fixed resource budget. They can modify training data, the trainer, and the inference harness using validation feedback before selecting a final system. The test makes resource allocation across interacting components part of the task. VibeTraining-Bench: Data-Trainer-Harness Co-Design for Iterative System Improvement.

Why this matters

Keep the constraints that govern real work inside the evaluation. A final outcome can hide a skipped obligation or an improvement bought with excessive resources.

Scan coverage

This is a keyword-filtered discovery feed, not a representative survey of AI research. The archived briefing used 177 of 755 captured evidence records. These findings summarize its cited releases; discovery dates do not necessarily equal original publication dates.

Data and briefing: 2026-10-20 archived snapshot.