Agent-Environment Alignment via Automated Interface Generation
Kaiming Liu, Xuanyu Lei, Ziyue Wang, Peng Li, Yang Liu · 2025
Read Agent-Environment Alignment via Automated Interface Generation by Kaiming Liu, Xuanyu Lei, Ziyue Wang, Peng Li, Yang Liu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-21055.txt.
Description
Kaiming Liu and co-authors fix a quiet mismatch between AI agents and the worlds they act in, in work from 2025. Agents reach their environments through interfaces that define actions and feedback, and an agent's expectations about its actions often diverge from what actually happens.
The team shows this misalignment measurably caps performance. Their framework, ALIGN, rewrites the interface automatically, enriching both the environment's static description and the feedback after each step. It wraps around existing systems without touching agent or environment code. Across household tasks, web navigation, and tool use, success rates rose, by up to about 46 percent in one benchmark. The generated interfaces also transferred to other agents and models.
Catalog record
| CREATOR | Kaiming Liu, Xuanyu Lei, Ziyue Wang, Peng Li, Yang Liu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 104 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-21055