王乐业_Toward Self-Improving Agents：构建可自我增强的Agent工程体系.pdf

1、A G E N T E N G I N E E R I N GToward Self-Toward Self-Improving AgentsImproving AgentsBuilding a Self-Enhancing Agent Engineering SystemFrom task correction to reusable skills,generated practice,and reliable evaluationPARADIGM SHIFTWhy Self-Improving Agents,and Why Self-Improving Agents,and Why Now

2、?Why Now?The Discussion Is ShiftingThe Discussion Is ShiftingFrom Can agents complete tasks?Can agents complete tasks?to Can agents improve over time?Can agents improve over time?The fundamental question in agent research has evolved from static capability to dynamic learning and adaptation.Recent S

3、ystems HighlightRecent Systems HighlightLearning loopsLearning loops continuous feedback cyclesSkill accumulationSkill accumulation building reusable capabilitiesPersistent adaptationPersistent adaptation memory across sessionsSelf-improvement is becoming a Self-improvement is becoming a core aspira

4、tioncore aspiration of agent systems of agent systemsThe Trend Is ClearThe Trend Is ClearStatic ExecutionContinuous ImprovementContinuous ImprovementAgent focus is shifting from one-time task completion to sustained capability enhancementCONCRETE SIGNALA Concrete Signal:Hermes AgentA Concrete Signal

5、:Hermes AgentHermes Agent is explicitly positioned as a Hermes Agent is explicitly positioned as a self-improving AI agentself-improving AI agentIts official description emphasizes built-in capabilities for continuous learning and adaptation from experience.Core Self-Improvement FeaturesCore Self-Im

6、provement FeaturesBuilt-in learning loopContinuous feedback integrationCreating skills from experienceAutomated skill extractionImproving skills during useRuntime refinementPersistent knowledgeCross-session memorySelf-Evolution ProjectA separate Hermes project explores automated evolution of skillss