A New AI Frameworks for Self-Organizing, Evolving AI Agents
Autonomous agents play a crucial role in advancing Artificial General Intelligence, enabling problem decomposition and tool orchestration through Large Language Models (LLMs). However, existing paradigms face a critical trade-off. On one hand, reusable fixed workflows require manual reconfiguration upon environmental changes; on the other hand, flexible reactive loops fail to distill reasoning progress into transferable structures. We introduce Hierarchical Variable Agent (HiVA), a novel framework modeling agentic workflows as self-organized graphs with the Semantic-Topological Evolution (STEV) algorithm, which optimizes hybrid semantic-topological spaces using textual gradients as discrete-domain surrogates for backpropagation. The iterative process comprises Multi-Armed Bandit-infused forward routing, diagnostic gradient generation from environmental feedback, and coordinated updates that co-evolve individual semantics and topology for collective optimization in unknown environments. Experiments on dialogue, coding, Long-context Q&A, mathematical, and agentic benchmarks demonstrate improvements of 5–10% in task accuracy and enhanced resource efficiency over existing baselines, establishing HiVA’s effectiveness in autonomous task execution.
Dynamic Self-Improvement Mechanisms
Dynamic self-improvement mechanisms enable agents to adapt through feedback. Existing approaches can be categorized into two paradigms: the first involves text-based optimization methods, including both single-agent semantic (Zhang et al. 2024) and multi-agent collaborative optimization (Liang, Xu, and Dong 2025); the second employs reinforcement learning approaches for agent improvement (Guo et al. 2025). In contrast, HiVA enables co-evolution of agent parameters and topology from a singleton, achieving effective test-time-scaling for diverse tasks and scenarios.
Hierarchies and Topological Optimization
Hierarchical structures and topological optimization enhance MAS efficiency. MetaGPT (Hong et al. 2024) and MASAI (Wadhwa et al. 2024) streamline workflows with modular designs. MASS (Zhang et al. 2025b) proposes a Multi-Agent System Search for prompt and topology optimization. G-designer (Zhang et al. 2025c) uses graph neural networks for topology optimization without semantic integration. HiVA’s Bayesian-driven hierarchical evolution enables better modeling of individuals in self-organized MAS.
A new research paper introduces a groundbreaking framework that allows AI agents to organize and evolve on their own, a significant step toward creating more autonomous and adaptable artificial intelligence.
The paper, titled “HiVA: Self-organized Hierarchical Variable Agent via Goal-driven Semantic-Topological Evolution,” addresses a fundamental challenge in current multi-agent AI systems. These systems often face a trade-off: they are either built with fixed, rigid workflows that require manual reprogramming to handle new tasks, or they use flexible, reactive loops that are less efficient and fail to learn from their experiences in a structured way.
Core Insights: The HiVA Framework
HiVA (Hierarchical Variable Agent) introduces a novel approach that enables a multi-agent system to evolve both its individual agent capabilities and its collaborative structure simultaneously, starting from a single agent and growing into a complex, self-organized team.
1. Co-evolution of Semantics and Topology
The central idea behind HiVA is that for an AI system to be truly autonomous, it must learn not only what each agent should do (its semantics) but also how these agents should interact and be organized (its topology). HiVA models the multi-agent system as a dynamic graph where each agent is a node. The system can change an agent’s internal instructions and tools (semantic evolution) and alter the connections between agents (topological evolution) based on feedback.
2. Learning from Experience with “Textual Gradients”
Since the interactions between language-based agents are not suitable for traditional optimization methods like backpropagation, HiVA uses a novel technique called “textual gradients.” When the system receives feedback from its environment (for example, a wrong answer or a failed task), it uses a Large Language Model (LLM) to translate that feedback into structured, actionable critiques. This “textual gradient” guides the system on what to change, whether it’s refining an agent’s prompt or restructuring the team.
3. Dynamic and Efficient Routing
HiVA doesn’t activate every agent for every task. It employs a smart routing mechanism called Knowledge-Aware Bayesian-Bandit (KABB) to dynamically select the most relevant agents for a specific task and construct a temporary “execution subgraph.” This makes the system more efficient and directs the problem to the agents best equipped to handle it.
4. The Agent Structure Itself is a Form of Memory
In the HiVA framework, the structure of the agent network serves as a form of long-term memory. As agents collaborate on tasks, successful interaction pathways are strengthened, while inefficient ones are pruned. This allows the system to remember and reuse effective problem-solving structures over time.
Performance and a Step Forward
The researchers tested HiVA across a wide range of benchmarks, including mathematical reasoning, coding, and long-context question answering. The results showed that HiVA improved task accuracy by 5–10% and enhanced resource efficiency compared to existing baseline models.
By enabling AI systems to autonomously evolve their skills and collaborative structures, the HiVA framework represents a significant step towards creating more general-purpose, adaptive, and efficient AI agents capable of tackling complex, real-world challenges.
Repurposing large vision-language models (LVLMs) as computer use agents (CUAs) has led to substantial breakthroughs, primarily driven by human-labeled data. However, these models often struggle with novel and specialized software, particularly in scenarios lacking human annotations. To address this challenge, we propose SEAgent, an agentic self-evolving framework enabling CUAs to autonomously evolve through interactions with unfamiliar software. Specifically, SEAgent empowers computer-use agents to autonomously master novel software environments via experiential learning, where agents explore new software, learn through iterative trial-and-error, and progressively tackle auto-generated tasks organized from simple to complex. To achieve this goal, we design a World State Model for step-wise trajectory assessment, along with a Curriculum Generator that generates increasingly diverse and challenging tasks. The agent’s policy is updated through experiential learning, comprised of adversarial imitation of failure actions and Group Relative Policy Optimization (GRPO) on successful ones. Furthermore, we introduce a specialist-to-generalist training strategy that integrates individual experiential insights from specialist agents, facilitating the development of a stronger generalist CUA capable of continuous autonomous evolution. This unified agent ultimately achieves performance surpassing ensembles of individual specialist agents on their specialized software. We validate the effectiveness of SEAgent across five novel software environments within OS-World. Our approach achieves a significant improvement of 23.2% in success rate, from 11.3% to 34.5%, over a competitive open-source CUA, i.e., UI-TARS.
Vision Language Models (VLMs) have recently achieved significant progress in bridging visual perception and linguistic reasoning. Recently, OpenAI o3 model introduced a zoom-in search strategy that effectively elicits active perception capabilities in VLMs, improving downstream task performance. However, enabling VLMs to reason effectively over appropriate image regions remains a core challenge in GUI grounding, particularly under high-resolution inputs and complex multi-element visual interactions. In this work, we propose LASER, a self-evolving framework that progressively endows VLMs with multi-step perception capabilities, enabling precise coordinate prediction. Specifically, our approach integrate Monte Carlo quality estimation with Intersection-over-Union (IoU)-based region quality evaluation to jointly encourage both accuracy and diversity in constructing high-quality preference data. This combination explicitly guides the model to focus on instruction-relevant key regions while adaptively allocating reasoning steps based on task complexity. Comprehensive experiments on the ScreenSpot Pro and ScreenSpot-v2 benchmarks demonstrate consistent performance gains, validating the effectiveness of our method. Furthermore, when fine-tuned on GTA1–7B, LASER achieves a score of 55.7 on the ScreenSpot-Pro benchmark, establishing a new state-of-the-art (SoTA) among 7B-scale models.
AI agents represent a shift from reactive to proactive automation. Unlike traditional software that waits for input, agents: • Maintain persistent memory across interactions • Execute tasks independently in your systems • Operate continuously without human oversight • Adapt their approach based on outcomes The key difference: agents complete workflows, not just conversations.
As AI advances toward general intelligence, the focus is shifting from systems optimized for static tasks to creating open-ended agents that learn continuously and adapt autonomously. This vision prioritizes long-term memory, skill transfer, and strategic planning, driven by an intrinsic curiosity to learn in dynamic, unpredictable environments. In this paper, we introduce Experience-driven Lifelong Learning (ELL), a framework for building self-evolving agents capable of continuous growth through real-world interaction. The framework is built on four core principles: (1) Experience Exploration: Agents learn through self-motivated interaction with dynamic environments, navigating interdependent tasks and generating rich experiential trajectories. (2) Long-term Memory: Agents preserve and structure historical knowledge, including personal experiences, domain expertise, and commonsense reasoning, into a persistent memory system. (3) Skill Learning: Agents autonomously improve by abstracting recurring patterns from experience into reusable skills, which are actively refined and validated for application in new tasks. (4) Knowledge Internalization: Agents internalize explicit and discrete experiences into implicit and intuitive capabilities as “second nature”. We also introduce StuLife, a benchmark dataset for ELL that simulates a student’s holistic college journey, from enrollment to academic and personal development, across three core phases and ten detailed sub-scenarios. StuLife is designed around three key paradigm shifts: From Passive to Proactive, From Context to Memory, and From Imitation to Learning. In this dynamic environment, agents must acquire and distill practical skills and maintain persistent memory to make decisions based on evolving state variables (e.g., resource availability and time). Critically, these agents are also expected to demonstrate intrinsic motivation by setting their own goals and initiating actions without external prompting. StuLife provides a comprehensive platform for evaluating lifelong learning capabilities, including memory retention, skill transfer, and self-motivated behavior. Beyond evaluating state-of-the-art LLMs on the StuLife benchmark, we also explore the role of context engineering in advancing AGI. Our results suggest that optimizing how we guide models may be as crucial as improving the models themselves, positioning context engineering as a key enabler of progress toward AGI
In this paper, we introduced Experience-driven Lifelong Learning (ELL), a framework for building self-evolving AI agents that learn and grow through continuous interaction with dynamic environments. Inspired by human cognitive development, ELL emphasizes three core principles: learning from experience, long-term memory integration, and skill abstraction, enabling agents not only to adapt to change but to proactively shape their own learning trajectories. To advance research in this direction, we present StuLife, a comprehensive benchmark that simulates a student’s college journey as a rich, longitudinal environment with interconnected academic, social, and planning tasks. StuLife goes beyond conventional benchmarks by incorporating evolving contexts, sparse rewards, and the need for selfinitiated behavior, making it a powerful platform for evaluating lifelong learning, memory retention, and autonomous decision-making. Our experiments reveal critical limitations in current LLM-based agents — particularly their lack of self-motivation and poor long-term memory — highlighting the gap between today’s systems and truly self-evolving intelligence. Importantly, we demonstrate that performance on complex, extended tasks can be significantly improved not only through model scale but through context engineering, particularly via well-designed system prompts. This suggests that the path to AGI may depend as much on how we structure an agent’s cognitive framework as on the raw capabilities of the underlying model. Looking forward, ELL and StuLife provide a principled foundation for developing agents that accumulate knowledge, transfer skills, and evolve autonomously over time. By shifting the focus from isolated task performance to sustained, experience-driven growth, we aim to catalyze progress toward intelligent systems that, like humans, become wiser with every experience.
Large language models (LLMs) often have a fixed knowledge cutoff, limiting their accuracy on emerging information. We present ALAS (Autonomous Learning Agent System), a modular pipeline that continuously updates an LLM’s knowledge with minimal human intervention. ALAS autonomously generates a learning curriculum for a target domain, retrieves up-to-date information from the web (with citations), distills this into question-answer training data, and fine-tunes the model through supervised fine-tuning (SFT) and direct preference optimization (DPO). It iteratively evaluates performance and revises the curriculum, enabling long-term continual learning. We demonstrate ALAS’s ability to self-improve a model on rapidly evolving domains (e.g., new Python releases, latest security CVEs, academic trends), significantly boosting post-cutoff question answering accuracy (from 15% to 90% on average) without manual dataset curation. The system emphasizes modularity and reproducibility: each component (planning, retrieval, distillation, memory, fine-tuning) is interchangeable and built on standard APIs. We discuss comparative baselines (e.g., retrieval-augmented generation vs. fine-tuning) and show that ALAS achieves 85–90% accuracy on knowledge-updated queries with minimal engineering overhead. Finally, we outline limitations (cost, dependency on source quality) and future directions for autonomous lifelong learning in LLMs.
Project Repository: https://github.com/DhruvAtreja/ALAS
A pattern collection and sharing system for Claude Code users. Rather than a heavy framework, this is a lightweight toolkit for capturing and sharing successful development patterns that emerge naturally from real projects.
✨ Key Features
🧠 Self-Learning System — Automatically detects patterns and learns from every interaction
📊 Scalable Architecture — Handles massive codebases without context overflow
🤖 Multi-Agent Support — Specialized agents for different tasks (security, performance, testing)
📚 Agent-Specific RAG — Each agent maintains its own knowledge base
🔄 Automatic Evolution — System improves and evolves based on usage patterns
📋 Task Management — Built-in backlog system with automatic prioritization
Recent advancements in Large Language Models (LLMs) have demonstrated their potential in specialized fields such as biomedical research. However, their performance is often limited by the lack of domain-specific knowledge and reasoning abilities. To address this, we introduce STELLA, a Self-Evolving LLM Agent designed for complex biomedical research tasks. STELLA leverages a novel self-evolving mechanism that enables it to continuously learn and adapt from new data and experiences. Our agent architecture integrates a dynamic knowledge base, a reasoning module, and a self-correction component, allowing it to perform intricate tasks such as literature analysis, experimental design, and data interpretation. We evaluate STELLA on a comprehensive benchmark of biomedical question-answering and research simulation tasks, demonstrating its superior performance over existing state-of-the-art models. Our results highlight the potential of self-evolving agents to accelerate biomedical discovery and innovation.
