Adaptive Explanation, Conditional Plans, and Plan Reuse

Talking Shop
1

Hello. For the use case of adaptive explanation, answering users’ questions and participating in explanatory dialogues with users while making use of dynamic user models, I am exploring conditional planning, where agents’ plans could include choice-points, where routes through conditional plans could be selected based upon observations made during plan execution, e.g., environments, dynamic user models, or state data.

Dynamic user modeling is an intricate topic. From a previous discussion, I am exploring layered approaches using LangGraph. With respect to user modeling (a.k.a., “partner modeling”), there are, to consider, both intra-session factors (affect, attentiveness, cooperativeness, cognitive load) and inter-session factors (preferences, interests, background knowledge, terminology knowledge, expertise).

For discussion, with respect to patterns like “plan-and-execute” and “plan-execute-validate”, it seems to me that, in addition to generating plans contextually and in an on-the-fly manner, agents could store, merge or combine, index, search for, retrieve, load, and reuse conditional plans.

In addition to reducing computational costs by reusing plans, optimistically, agents could utilize pooled collective knowledge to learn from their bulk interactions with explainees.

Is anyone else here in the community interested in adaptive explanation at scale? I’ve attached a preliminary bibliography. Do any other preprints or publications come to mind to share or recommend?

Are there any open-source examples or projects which involve conditional planning (generating and executing plans containing choice-points) and/or plan reuse, i.e., storing and loading plans from repositories or libraries of plans?

Thank you.

Bibliography

Booshehri, Meisam, Hendrik Buschmeier, Milad Alshomary, Katharina Rohlfing, Henning Wachsmuth, and Philipp Cimiano. “Modeling explanations as processes: An analytical framework accounting for relational and structural patterns in explanatory dialogues.” (2024).

Cawsey, Alison. Explanation and Interaction: The Computer Generation of Explanatory Dialogues. MIT Press, 1992.

Fichtel, Leandra, Maximilian Spliethöver, Eyke Hüllermeier, Patricia Jimenez, Nils Klowait, Stefan Kopp, Axel-Cyrille Ngonga Ngomo, Amelie Robrecht, Ingrid Scharlau, Lutz Terfloth, Anna-Lisa Vollmer, and Henning Wachsmuth. “Investigating co-constructive behavior of large language models in explanation dialogues.” In Proceedings of the 26th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pp. 1-20. 2025.

Mindlin, Dimitry, Meisam Booshehri, and Philipp Cimiano. “Towards co-constructed explanations: A multi-agent reasoning-based conversational system for adaptive explanations.” In Proceedings of the 13th International Conference on Human-Agent Interaction, pp. 148-157. 2025.

Moore, Johanna D. Participating in Explanatory Dialogues: Interpreting and Responding to Questions in Context. MIT Press, 1994.

Robrecht-Hilbig, Amelie S., Christoph Kowalski, and Stefan Kopp. “Generation and evaluation of adaptive explanations based on dynamic partner-modeling and non-stationary decision making.” Frontiers in Computer Science 8 (2026): 1558674.

Robrecht-Hilbig, Amelie S. “Adaptive explanations as co‐constructed processes: Modeling a rational explainer through the interaction of dynamic partner models and non‐stationary decision making.” (2026).

2

Along these lines, I recently found:

LLM-based agent applications have shown increasingly remarkable capabilities in complex workflows but incur substantial costs and latency due to extensive planning and reasoning requirements. Existing LLM caching techniques (like context caching and semantic caching), primarily designed for serving chatbots, are insufficient for agent applications where outputs depend on external data and environmental contexts. We propose Agentic Plan Caching (APC), a novel test time memory that extracts, stores, adapts, and reuses structured plan templates from planning stages of agent applications across semantically similar tasks to reduce the cost and latency of serving. Unlike traditional semantic caching, our system extracts plan templates from completed agent executions at test-time, employs keyword extraction to match new requests against cached plans, and utilizes lightweight models to adapt these templates to task-specific plans with contexts. Evaluation across multiple real-world agent applications shows that our system can reduce costs by 50.31% and latency by 27.28% on average while maintaining performance, offering a more efficient solution for serving LLM-based agents that complements existing LLM serving infrastructures.

Zhang, Qizheng, Michael Wornow, and Kunle Olukotun. “Agentic plan caching: Test-time memory for fast and cost-efficient LLM agents.” Advances in Neural Information Processing Systems 38 (2026): 103270-103296.

3

With respect to LLM-based agents and case-based reasoning (CBR) and planning (CBP), I found Memento.

To this end, we introduce Memento, a non-parametric, learn-on-the-fly framework for CBR (Smyth and McClave, 2001, Hatalis et al., 2025), instantiated as a planner–executor architecture grounded in a memory based Markov Decision Process (MDP). Memento comprises three principal components: (i) a planner, (ii) a tool-enabled executor, and (iii) a growing Case Bank that stores past trajectories as episodic memory. Instead of relying solely on the LLM’s parametric memory, which is fixed after training, online case-based reasoning in Memento is implemented by storing rich episodic traces.

Zhou, Huichi, Yihang Chen, Siyuan Guo, Xue Yan, Kin Hei Lee, Zihan Wang, Ka Yiu Lee et al. “Memento: Fine-tuning LLM agents without fine-tuning LLMs.” arXiv preprint arXiv:2508.16153 (2025).