Can Large Language Models Adapt to Other Agents In-Context?