Selected Work
-
Better, Faster, Stronger: Programmatic Skill Learning Best Reduces Agent Cost
Zixi Huang,* Xiheng Wang,* Andrew Wang*, William Jurayj, Bernal Jiménez Gutiérrez, Daniel Khashabi, Nicholas Andrews
Arxiv preprint, 2026Agents should become more efficient as they gain more task experience. To that end, we argue that the modality of code is an effective representation of experience. Codifying common behavior into programs avoids calling an LLM to re-reason every step. We show that across models and embodied tasks, our approach decreases cost more than text-based skill memory while matching performance.
-
Hell or High Water: Evaluating Agentic Recovery from External Failures
Andrew Wang, Sophia Hager, Adi Asija, Daniel Khashabi, Nicholas Andrews.
COLM, 2025How good are agents at coming up with backup plans when the tools they rely on stop working? We create a benchmark where each task is solvable with two different sets of tools, so that one broken tool never makes the task impossible. We find that LLMs struggle in this setting, and more available tools makes their performance even worse.
Papers
-
Better, Faster, Stronger: Programmatic Skill Learning Best Reduces Agent Cost
-
AgentOdyssey: Open-Ended Long-Horizon Text Game Generation for Test-Time Continual Learning Agents
-
Hell or High Water: Evaluating Agentic Recovery from External Failures
-
Feedback Friction: LLMs Struggle to Fully Incorporate External Feedback
-
Learning Extrapolative Sequence Transformations from Markov Chains
-
RATIONALYST: Pre-training Process-Supervision for Improving Reasoning
-
AnaloBench: Benchmarking the Identification of Abstract and Long-context Analogies
-
Learning to Generate Text in Arbitrary Writing Styles
-
Can Authorship Representation Learning Capture Stylistic Features?
* equal contribution
Talks
-
2024 JSALT (Annual Frederick Jelinek Memorial Summer Workshop)
AnaloBench: Benchmarking the Identification of Abstract and Long-context Analogies -
EMNLP 2023
Can Authorship Representation Learning Capture Stylistic Features?
Updated 2026/09/22
