AdaptAgent: Adapting Multimodal Web Agents with Few-Shot Learning from Human Demonstrations

Gaurav Verma , Rachneet Kaur , Nishan Srishankar , Zhen Zeng , Tucker Balch , Manuela Veloso

🏛 Institutions: Georgia Tech , J.P. Morgan AI Research
📅 Date: November 24, 2024
📑 Publisher: ACL 2025
💻 Env: Web
🔑 Keywords: framework few-shot adaptation human demonstrations website adaptation AdaptAgent

TLDR

AdaptAgent studies how multimodal web agents can adapt to unseen websites with only a few human demonstrations instead of relying solely on broad pretraining or large-scale fine-tuning. It shows that both proprietary and open-weight agents benefit from few-shot demonstrations, with clear gains on Mind2Web and VisualWebArena.

Open paper Report issue