No items found.

Adaptive Environments For LLM Agents Need to Jointly Scale Task, Scaffolding and Reward Design

Darshan Deshpande, Varun Gangal, Li Siyan, Anand Kannappan, Rebecca Qian

Darshan Deshpande*, Varun Gangal*, Li Siyan*, Anand Kannappan, Rebecca Qian

*Equal contribution

“A Study in Color”, Kenya
03

Abstract

As machine learning systems become increasingly agentic, static dataset mixes, hand-written environments and arenas, and fixed benchmarks are proving insufficient for post-training and evaluation beyond a certain extent, as continuous training or hill-climbing through repeated evaluation is bound to saturate them. This limitation persists even for interactive benchmarks with fixed configurations. Recent work, therefore, argues for dynamic benchmarks that co-evolve with model capabilities, but typically scales only one environmental component at a time (the tasks, tools, or reward design). In this position paper, we showcase the need for jointly scaling all three components as a more robust foundation for dynamic evaluation and training. Our position builds on curriculum learning and meta-evaluation theory through which we sketch a class of joint-scaling algorithms and present a GridWorld thought experiment showing that scaling any single component can hit fundamental limits. We recommend that future environment-centric benchmarking and post-training methods be designed with this joint-scaling paradigm in consideration.

03

Runs

Cite this
TextBibTeXCopy
Preprint
February 18, 2026
generative-simulators
More research
All research →
“Earth’s Aquarium”, Bering Sea“Earth’s Aquarium”, Bering Sea
NeurIPS 2026
meta

Masked Diffusion Language Models are Strong and Steerable Text-Based World Models for Agentic RL

May 7, 2026
NeurIPS 2026
Main
world-modeling-for-agentic-rl
NeurIPS 2026
“Konari, Iran”, Zagros Mountains“Konari, Iran”, Zagros Mountains
GEM Workshop at ACL 2026
meta

DETOUR: An Interactive Benchmark for Dual-Agent Search and Reasoning

January 30, 2026
ACL 2026 workshop
Workshop
detour
GEM Workshop at ACL 2026
“Himalayas”, Eastern Himalaya, southwestern China“Himalayas”, Eastern Himalaya, southwestern China
Nature
meta

Humanity’s Last Exam

January 24, 2025
Nature 2026
Main
humanitys-last-exam
Nature
From the blog
All posts →