# Efficient Test-Time Adaptation through Human-Agent Interaction

DevFeed: [Efficient Test-Time Adaptation through Human-Agent Interaction](<https://devfeed.tech/articles/efficient-test-time-adaptation-through-human-agent-interaction-135004.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.04141>)

Author: Zora Zhiruo Wang, Apurva Gandhi, Rulin Shao, Aspen Chen, Jonas Mueller, Zhiqi Liang, Jett Chen, Michael Ryan, Qianou Ma, Luxi He, Zhoujun Cheng, Andre He, Seungone Kim, Jiayi Geng, Mingqian Zheng, Wei

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [Test automation](<https://devfeed.tech/topics/test-automation.md>), [human computer interaction](<https://devfeed.tech/topics/human-computer-interaction.md>), [knowledge-engineering](<https://devfeed.tech/topics/knowledge-engineering.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agent-context](<https://devfeed.tech/tags/agent-context.md>), [ai](<https://devfeed.tech/tags/ai.md>), [annotation](<https://devfeed.tech/tags/annotation.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [failures](<https://devfeed.tech/tags/failures.md>), [humans](<https://devfeed.tech/tags/humans.md>), [improvements](<https://devfeed.tech/tags/improvements.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [module](<https://devfeed.tech/tags/module.md>), [need](<https://devfeed.tech/tags/need.md>), [open](<https://devfeed.tech/tags/open.md>), [personal](<https://devfeed.tech/tags/personal.md>), [rich](<https://devfeed.tech/tags/rich.md>), [scale](<https://devfeed.tech/tags/scale.md>), [session](<https://devfeed.tech/tags/session.md>), [signal](<https://devfeed.tech/tags/signal.md>), [solo](<https://devfeed.tech/tags/solo.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [test](<https://devfeed.tech/tags/test.md>)

## AI overview

TAHI uses cross-session human-agent interactions to adapt agents to individual preferences, incorporating feedback into agent context and weights while an evolving rubric module captures each user's training and evaluation criteria. In experiments with 30 people across writing and visual creation (600 tasks), adaptation improved solo task success by 4.5-20.9% within tens of tasks. The rubrics identified 16.0-22.3% more failures than rubrics from language models or humans alone. Personalized agents also showed success improvements of up to 8.8% that generalized across users.

## Source excerpt

arXiv:2609.04141v2 Announce Type: replace Abstract: AI agents are trained on population-scale data to encode broad capabilities spanning those of many practitioners. Yet the artifacts they produce rarely meet the personal bar professionals need to stake their reputation on. On realistic, open-ended tasks where success criteria are heterogeneous and insufficiently documented, individual expertise lives precisely in the elevation and departure from the average. In practice, iterative human-agent interaction surfaces criteria that users cannot fully specify up front, yet apply repeatedly across tasks. We argue this cross-session interaction data is a rich, underused signal for closing the gap to individual expertise. In this work, we propose test-time adaptation through human-agent interaction (TAHI), which integrates these signals into agent context and weights, and crystallizes each user's training and evaluation criteria via an evolving rubric module. We adapt agents to 30 individuals in two high-utility domains, writing and visual creation, on a total of 600 tasks. Our agents improve solo task success by 4.5-20.9% within only tens of tasks. Meanwhile, our evolving rubric module serves as a scalable annotation tool, creating evaluation rubrics that catch 16.0-22.3% more failures than those from LMs or humans alone. While agents are adapted towards individuals, we show these personalized agents also produce improvements in success of up to 8.8% that generalize across users.