AgentCL: Toward Rigorous Evaluation of Continual Learning in Language Agents
Summary
A comprehensive evaluation framework for continual learning in language agents is introduced, emphasizing controlled task streams and memory design analysis to better assess reusable experience and learning stability.
View Cached Full Text
Cached at: 06/04/26, 03:42 AM
Paper page - AgentCL: Toward Rigorous Evaluation of Continual Learning in Language Agents
Source: https://huggingface.co/papers/2606.02461
Abstract
A comprehensive evaluation framework for continual learning in language agents is introduced, emphasizing controlled task streams and memory design analysis to better assess reusable experience and learning stability.
Language agentsspend substantial inference time solving individual tasks, yet the experience acquired in one episode is often underutilized in future episodes.Continual learningexpects an agent to accumulate reusable experience across a stream of tasks, improve over time, and avoid interference from irrelevant experiences. Unfortunately, existing benchmarks struggle to evaluatecontinual learninginlanguage agentsrigorously. Most efforts focus on retrieval and reasoning over long-context conversations or documents, while recent lifelong-adaptation benchmarks often rely on naivetask streamswith limited analysis of cross-task relationships, making it difficult to understand what an agent learns and reuses over time. This paper presents an evaluation framework AgentCL forcontinual learningin agents, centered on controlledtask streamsand metrics fortransfer gains. AgentCL constructscompositional streamswhere earlier sub-solutions, evidence, or workflows are intentionally reusable in later tasks, and contrasts them with naive streams where such reusability is not guaranteed. We use the benchmark to evaluatenon-parametric memorydesigns forcontinual learning. To diagnose how memory design choices affectcontinual learning, we developMemProbe, a probing method that stores interactions, insights, and skills, while filtering unreliable experiences during consolidation. Empirical analysis across coding, deep research, and language understanding/reasoning tasks shows that naive streams offer limited ability to distinguishmemory designs, whereas controlled streams more clearly distinguish theirplasticity. Meanwhile, naive and held-out settings often yield limited gains and can expose memory-induced degradation. These results highlight the need for strongermemory designsthat balanceplasticityandstable reuse.
View arXiv pageView PDFAdd to collection
Get this paper in your agent:
hf papers read 2606\.02461
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2606.02461 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2606.02461 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2606.02461 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
@omarsar0: // Continual Learning Bench // One of the research areas with lots of investments is continual learning. While there ar…
CL-Bench is a new expert-validated benchmark across six domains that evaluates whether LLM-based agents genuinely learn from sequential experience. It finds that naive in-context learning often outperforms dedicated memory systems, indicating current architectures add overhead rather than genuine learning.
Rethinking Continual Experience Internalization for Self-Evolving LLM Agents
This paper investigates why LLM agents suffer from progressive capability collapse under multi-iteration experience internalization and proposes a robust recipe addressing experience granularity, injection patterns, and training regime. Key findings include that principle-level experience, step-wise injection, and off-policy context-distillation yield more stable and sustainable continual learning.
CLaaS: Continual learning as a service for sample efficient online learning
CLaaS is a system for continual learning of LLM agents in deployment, using experience replay for sample-efficient online adaptation.
Do Agent Optimizers Compound? A Continual-Learning Evaluation on Terminal-Bench 2.0
This paper introduces a two-phase continual-learning evaluation on Terminal-Bench 2.0 to test whether gains from agent-optimization methods compound when applied recursively. It finds that only RELAI-VCL, which incorporates regression control, achieves compounded improvement.
Continual Learning for Agents (3 minute read)
Replit built ViBench to evaluate app-building success from natural-language specs and Telescope to cluster production failure traces, enabling harness-level and context-level continual learning for agents using closed frontier models.