The rapid progress of foundation models has produced agents with remarkable capabilities in perception, language understanding, reasoning, and tool use, while advances in post-training, reinforcement learning, retrieval-augmented generation, and agentic scaffolding now let both virtual and embodied agents tackle complex tasks in coding, web navigation, scientific discovery, and robotic control. Yet today's agents remain largely static: after costly pre- and post-training, their knowledge, skills, and behaviors are mostly fixed, and once deployed they adapt through prompting, retrieval, or external tools rather than genuine internal learning and memory consolidation.
This contrasts sharply with human intelligence, where people continuously acquire knowledge, refine representations, and reorganize beliefs through interaction. Current agents, whether virtual or embodied, instead cannot continually learn at test time: they struggle to internalize new information after deployment, fail to improve from repeated mistakes, and can forget prior knowledge and skills when updated naively (catastrophic forgetting).
The NeurIPS 2026 Workshop on Towards Test-Time Continual Learning Agents (TTCL) brings together researchers across continual learning, large language models, reinforcement learning, embodied AI, memory systems, cognitive science, robotics, and multimodal learning. We define Test-Time Continual Learning Agents as AI systems that continuously acquire, consolidate, and refine knowledge and capabilities during deployment, without catastrophic forgetting or repeated large-scale retraining. This goes beyond updating facts: it asks how agents improve perception, reasoning, planning, exploration, skill acquisition, and long-term decision-making through ongoing experience in virtual and physical worlds. These capabilities are especially important for robotics, scientific discovery, personalized assistants, education, healthcare, and human–AI collaboration.
