Recursive Experiential-Working Memory Evolution for Long-Horizon Agent Harnesses
2026-08-25 • Artificial Intelligence
Artificial IntelligenceComputation and Language
AI summaryⓘ
The authors introduce Recuris, a new system designed to help AI agents improve their performance on long and complex tasks by using a special memory setup. This system keeps track of what the agent has done so far and helps it pick the right skills based on the current situation instead of confusing old history. It also improves by learning from mistakes in a step-by-step way. In tests, Recuris made several AI models much better at completing difficult tasks, especially as tasks got longer and harder. The authors suggest that this method can help AI agents get better by building on their past experiences in a smart, organized way.
Recursive self-improvementExperiential MemoryWorking MemoryLong-horizon tasksSkill selectionMeta-AgentMemory-evolution loopTask successAI agent architectureValidation-gated updates
Authors
Zhaochen Yu, Yingcheng Wu, Zhenfei Yin, Kaiyuan Chen, Zhe Zhao, Mengdi Wang, Shuicheng Yan, Ling Yang
Abstract
Recursive self-improvement (RSI) remains hard in long-horizon tasks, where growing histories obscure the task state and misalign skill invocation. We introduce Recuris, a recursive Experiential-Working Memory architecture for long-horizon agent harnesses, in which Working Memory tracks task progress and guides skill selection from Experiential Memory, grounding skill use in current needs rather than the full history. This coupling also turns execution into structured evidence that localizes failures to specific memory components. Across tasks, a fixed Meta-Agent turns that evidence into localized, validation-gated updates to Skill Memory that reshape execution and yield new evidence, forming a bounded recursive memory-evolution loop. Across four long-horizon benchmarks and ten models, Recuris improves task success in 35 of the 37 completed model-benchmark pairs, carrying frontier models to SOTA-level task success: on tau-bench it adds +17.8 points to GPT-5.6 Sol and +15.6 to Claude Opus 5, taking Opus 5 to 87.9%, and +16.6/+13.5 points on Qwen3.6-27B/35B on SkillFlow. The advantage widens as the interaction horizon grows, to +32.2 points on the longest tasks, and common long-horizon failures fall by up to 80%. These results position recursively evolving memory as a scalable foundation for RSI, enabling agents to continuously transform accumulated experience into increasingly effective long-horizon behavior. Code: https://github.com/Gen-Verse/Recuris