Adaptation of Agentic AI: A Survey of Post-Training, Memory, and Skills
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917323976212480 |
|---|---|
| author | Jiang, Pengcheng Lin, Jiacheng Shi, Zhiyi Wang, Zifeng He, Luxi Wu, Yichen Zhong, Ming Song, Peiyang Zhang, Qizheng Wang, Heng Xu, Xueqiang Xu, Hanwen Han, Pengrui Zhang, Dylan Sun, Jiashuo Yang, Chaoqi Qian, Kun Wang, Tian Hu, Changran Li, Manling Li, Quanzheng Peng, Hao Wang, Sheng Shang, Jingbo Zhang, Chao You, Jiaxuan Liu, Liyuan Lu, Pan Zhang, Yu Ji, Heng Choi, Yejin Song, Dawn Sun, Jimeng Han, Jiawei |
| author_facet | Jiang, Pengcheng Lin, Jiacheng Shi, Zhiyi Wang, Zifeng He, Luxi Wu, Yichen Zhong, Ming Song, Peiyang Zhang, Qizheng Wang, Heng Xu, Xueqiang Xu, Hanwen Han, Pengrui Zhang, Dylan Sun, Jiashuo Yang, Chaoqi Qian, Kun Wang, Tian Hu, Changran Li, Manling Li, Quanzheng Peng, Hao Wang, Sheng Shang, Jingbo Zhang, Chao You, Jiaxuan Liu, Liyuan Lu, Pan Zhang, Yu Ji, Heng Choi, Yejin Song, Dawn Sun, Jimeng Han, Jiawei |
| contents | Large language model (LLM) agents are moving beyond prompting alone. ChatGPT marked the rise of general-purpose LLM assistants, DeepSeek showed that on-policy reinforcement learning with verifiable rewards can improve reasoning and tool use, and OpenClaw highlights a newer direction in which agents accumulate persistent memory and reusable skills. Yet the research landscape remains fragmented across post-training, retrieval, memory, and skill systems. This survey studies these developments under a single notion of \emph{adaptation}: improving an agent, its tools, or their interaction after pretraining. We organize the field with a four-paradigm framework spanning agent adaptation and tool adaptation. On the agent side, A1 (tool-execution-signaled) and A2 (agent-output-signaled) improve the agent itself through supervised fine-tuning, preference optimization, and reinforcement learning with verifiable rewards. On the tool side, T1 (agent-agnostic) provides reusable pre-trained modules any agent can call, while T2 (agent-supervised) uses the agent's outputs to train memory systems, skill libraries, or lightweight subagents. Using this framework, we review post-training methods, adaptive memory architectures, and agent skills; compare their trade-offs in cost, flexibility, and generalization; and summarize evaluation practices across deep research, software development, computer use, and drug discovery. We conclude by outlining open problems in agent-tool co-adaptation, continual learning, safety, and efficient deployment. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_16301 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Adaptation of Agentic AI: A Survey of Post-Training, Memory, and Skills Jiang, Pengcheng Lin, Jiacheng Shi, Zhiyi Wang, Zifeng He, Luxi Wu, Yichen Zhong, Ming Song, Peiyang Zhang, Qizheng Wang, Heng Xu, Xueqiang Xu, Hanwen Han, Pengrui Zhang, Dylan Sun, Jiashuo Yang, Chaoqi Qian, Kun Wang, Tian Hu, Changran Li, Manling Li, Quanzheng Peng, Hao Wang, Sheng Shang, Jingbo Zhang, Chao You, Jiaxuan Liu, Liyuan Lu, Pan Zhang, Yu Ji, Heng Choi, Yejin Song, Dawn Sun, Jimeng Han, Jiawei Artificial Intelligence Computation and Language Large language model (LLM) agents are moving beyond prompting alone. ChatGPT marked the rise of general-purpose LLM assistants, DeepSeek showed that on-policy reinforcement learning with verifiable rewards can improve reasoning and tool use, and OpenClaw highlights a newer direction in which agents accumulate persistent memory and reusable skills. Yet the research landscape remains fragmented across post-training, retrieval, memory, and skill systems. This survey studies these developments under a single notion of \emph{adaptation}: improving an agent, its tools, or their interaction after pretraining. We organize the field with a four-paradigm framework spanning agent adaptation and tool adaptation. On the agent side, A1 (tool-execution-signaled) and A2 (agent-output-signaled) improve the agent itself through supervised fine-tuning, preference optimization, and reinforcement learning with verifiable rewards. On the tool side, T1 (agent-agnostic) provides reusable pre-trained modules any agent can call, while T2 (agent-supervised) uses the agent's outputs to train memory systems, skill libraries, or lightweight subagents. Using this framework, we review post-training methods, adaptive memory architectures, and agent skills; compare their trade-offs in cost, flexibility, and generalization; and summarize evaluation practices across deep research, software development, computer use, and drug discovery. We conclude by outlining open problems in agent-tool co-adaptation, continual learning, safety, and efficient deployment. |
| title | Adaptation of Agentic AI: A Survey of Post-Training, Memory, and Skills |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2512.16301 |