The Hierarchy of Agentic Capabilities: Evaluating Frontier Models on Realistic RL Environments

TL;DR

This paper empirically evaluates leading AI models on 150 realistic workplace tasks, revealing a hierarchy of necessary agentic skills and highlighting significant performance gaps in complex, multi-step environments.

Contribution

It introduces a hierarchy of agentic capabilities for AI models and a task-centric RL environment design methodology, providing insights into current model limitations and development directions.

Findings

Models fail about 40% of tasks, with failures aligned to capability hierarchy.

Weaker models struggle with tool use and planning; stronger models falter on contextual inference.

Proposes a diverse, expert-informed RL environment design for better evaluation.

Abstract

The advancement of large language model (LLM) based agents has shifted AI evaluation from single-turn response assessment to multi-step task completion in interactive environments. We present an empirical study evaluating frontier AI models on 150 workplace tasks within a realistic e-commerce RL environment from Surge. Our analysis reveals an empirically-derived \emph{hierarchy of agentic capabilities} that models must master for real-world deployment: (1) tool use, (2) planning and goal formation, (3) adaptability, (4) groundedness, and (5) common-sense reasoning. Even the best-performing models fail approximately 40\% of the tasks, with failures clustering predictably along this hierarchy. Weaker models struggle with fundamental tool use and planning, whereas stronger models primarily fail on tasks requiring contextual inference beyond explicit instructions. We introduce a task-centric design methodology for RL environments that emphasizes diversity and domain expert contributions, provide detailed failure analysis, and discuss implications for agent development. Our findings suggest that while current frontier models can demonstrate coherent multi-step behavior, substantial capability gaps remain before achieving human-level task completion in realistic workplace settings.