A Comprehensive Survey of Self-Evolving AI Agents: A New Paradigm Bridging Foundation Models and Lifelong Agentic Systems

To address this limitation, recent research has explored agent evolution techniques that aim to automatically enhance agent systems based on interaction data and environmental feedback. This emerging direction lays the foundation for self-evolving AI agents, which bridge the static capabilities of foundation models with the continuous adaptability required by lifelong agentic systems. In this survey, we provide a comprehensive review of existing techniques for self-evolving agentic systems. Specifically, we first introduce a unified conceptual framework that abstracts the feedback loop underlying the design of self-evolving agentic systems. The framework highlights four key components: System inputs, Agent System, Environment, and Optimisers, serving as a foundation for understanding and comparing different strategies. Based on this framework, we systematically review a wide range of self-evolving techniques that target different components of the agent system, including foundation models, agent prompts, memory, tools, workflows, and communication mechanisms across agents.

Despite the notable progress in agent systems, most of them, whether single- or multi-agent, continue to rely extensively on manually designed configurations. Once deployed, these systems typically maintain static architectures and fixed functionalities. However, real-world environments are dynamic and continuously evolving, e.g., user intents shift, task requirements change, and external tools or information sources may vary over time. For instance, an agent assisting in customer service may need to handle newly introduced products, updated company policies, or unfamiliar user intents. Similarly, a scientific research assistant may be required to incorporate a newly published algorithm, or integrate a novel analysis tool. In such settings, manually reconfiguring the agent system is time-consuming, labour-intensive, and difficult to scale.

These challenges have motivated recent efforts to explore the new paradigm of Self-Evolving AI Agents, a novel class of agent systems capable of autonomous adaptation and continuous self-improvement, bridging foundation models with lifelong learning agentic systems.

Three Laws of Self-Evolving AI Agents

I. Endure (Safety Adaptation)

Self-evolving AI agents must maintain safety and stability during any modification;

II. Excel (Performance Preservation)

Subject to the First law, self-evolving AI agents must preserve or enhance existing task performance;

III. Evolve (Autonomous Evolution)

Subject to the First and Second law, self-evolving AI agents must be able to autonomously optimise their internal components in response to changing tasks, environments, or resources.

We characterise the emergence of self-evolving AI agents as part of a broader paradigm shift in the development of LLM-based systems. This shift spans from early-stage Model Offline Pretraining (MOP) and Model Online Adaptation (MOA), to more recent trends in Multi-Agent Orchestration (MAO), and ultimately, to Multi-Agent Self-Evolving (MASE). As summarised in Figure 1 and Table 1, each paradigm builds on the previous one, moving from a static, frozen foundation model to fully autonomous, self-evolving agentic systems.

• MOP (Model Offline Pretraining). The initial stage focuses on pretraining foundation models on large-scale, static corpora and then deploying them in a fixed, frozen state, without further adaptation.

• MOA (Model Online Adaptation). Building on MOP, this stage introduces post-deployment adaptation, where the foundation models can be updated through techniques such as supervised fine-tuning, lowrank adapters (Pfeiffer et al., 2021; Hu et al., 2022), or reinforcement learning from human feedback (RLHF) (Ouyang et al., 2022), using labels, ratings, or instruction prompts.

• MAO (Multi-Agent Orchestration). Extending beyond a single foundation model, this stage coordinates multiple LLM agents that communicate and collaborate via message exchange or debate prompts (Li et al., 2024g; Zhang et al., 2025h), to solve complex tasks without modifying the underlying model parameters.

• MASE (Multi-Agent Self-Evolving). Finally, MASE introduces a lifelong, self-evolving loop where a population of agents continually refines their prompts, memory, tool-use strategies and even their interaction patterns based on environmental feedback and meta-rewards (Novikov et al., 2025; Zhang et al., 2025i).

The evolution from MOP to MASE represents a fundamental shift in the development of LLM-based systems, from static, manually configured architectures to adaptive, data-driven systems that can evolve in response to changing requirements and environments. Self-evolving AI agents bridge the static capabilities of foundation models with the continuous adaptability required by lifelong agentic systems, offering a path toward more autonomous, resilient, and sustainable AI.

Despite self-evolving AI agents representing an ambitious vision for future AI systems, achieving this level of autonomy remains a long-term goal. Current systems are still far from exhibiting the full capabilities required for safe, robust and open-ended self-evolution. In practice, current progress towards this vision is achieved through agent evolution and optimisation techniques, which provide practical means for enabling agent systems to iteratively refine their components based on interaction data and environmental feedback, thereby enhancing their effectiveness in real-world tasks. Recent research has explored several key directions in this area. One line of work focuses on enhancing the underlying LLM itself to improve the core capabilities, such as planning (Qiao et al., 2024), reasoning (Zelikman et al., 2022; Tong et al., 2024), and tool use (Feng et al., 2025a). Another line of research targets the optimisation of auxiliary components within agent systems, including prompts (Xu et al., 2022; Prasad et al., 2023; Yang et al., 2024a; Wang et al., 2025i), tools (Yuan et al., 2025b; Qu et al., 2025), memory (Zhong et al., 2024; Lee et al., 2024d), and etc., allowing the agents to better generalise to new tasks and dynamic environments. Furthermore, in multi-agent systems, recent work investigates the optimisation of agent topologies and communication protocols (Bo et al., 2024; Chen et al., 2025h; Zhang et al., 2025j; Zhou et al., 2025a), aiming to identify agent structures that are best suited to the current task and improve the coordination and information sharing among agents.

Existing surveys on AI agents either focus on the general introduction of agent architectures and functionalities (Wang et al., 2024c; Guo et al., 2024c; Xi et al., 2025; Luo et al., 2025a; Liu et al., 2025a,c), or target specific components such as planning (Huang et al., 2024b), memory (Zhang et al., 2024d), collaboration mechanism (Tran et al., 2025), and evaluation (Yehudai et al., 2025). Other surveys investigate domain-specific applications of agents, such as operating system agents (Hu et al., 2025b) and healthcare agents (Sulis et al., 2023). While these surveys provide valuable insights into various aspects of agent systems, recent advances in agent self-evolution and continual adaptation have not been sufficiently covered, which corresponds to the capabilities of agents that are central to the development of lifelong, autonomous AI systems. This leaves a critical gap in the literature for researchers and practitioners seeking a holistic understanding of the new learning paradigm that underpins adaptive and self-evolving agentic systems.

Specifically, we introduce a unified conceptual framework that abstracts the feedback loop underlying the design of self-evolving agentic systems. This framework identifies four core components: System Inputs, Agent System, Environment, and Optimisers, highlighting the evolution loop of agent systems. Building on this framework, we systematically examine a wide range of evolution and optimisation techniques that target different components of the agent systems, including the LLM, prompts, memory, tools, workflow topologies, and communication mechanisms. Moreover, we also investigate domain-specific evolution strategies developed for specialised fields.

An AI agent is typically composed of multiple components that work together to enable autonomous decisionmaking and execution. The core component of an agent is the Foundation Model, most commonly an LLM2, which serves as the central reasoning engine responsible for interpreting instructions, generating plans, and producing actionable responses. In addition, there are also some supporting modules that enhance the agent’s ability in complex and dynamic environments:

(1) Perception Module. The perception module is responsible for acquiring and interpreting information from the environment (Li et al., 2024f). This includes processing textual inputs, audio signals, video frames, or other sensory-like data to build a representation suitable for reasoning.

(2) Planning Module. The planning module enables the agent to decompose complex tasks into actionable sub-tasks or sequences of operations and guide their execution across multiple steps (Huang et al., 2024b). This process facilitates hierarchical reasoning and ensures coherent task completion. One of the simplest forms of planning involves linear task decomposition, where a problem is broken down into multiple intermediate steps, and the LLM follows these steps to address the problem. This is exemplified by methods such as chain-of-thought prompting (Wei et al., 2022). Beyond static planning, more dynamic approaches interleave planning and execution in an iterative loop. For instance, the ReAct (Yao et al., 2023b) framework combines reasoning with actions, allowing the agent to revise its plans based on real-time feedback. In addition to linear planning, some methods adopt a branching strategy, where each step may lead to multiple possible continuations. Representative examples are Tree-of-Thought (Yao et al., 2023a) and Graph-of-Thought (Besta et al., 2024), which enable the agent to explore multiple reasoning paths.

(3) Memory Module. The memory module enables the agent to retain and recall past experience, enabling context-aware reasoning and long-term consistency. Broadly, memory can be categorised into short-term and long-term memory. Short-term memory typically stores the context and interactions generated during the execution of the current task. Once the task is completed, the short-term memory will be removed. In contrast, long-term memory persists over time and may store accumulated knowledge, past experiences, or reusable information across tasks. To access relevant long-term memory, many agent systems adopt a retrieval-augmented generation (RAG) module (Zhang et al., 2024d), where the agent retrieves relevant information from the memory and incorporates them into the input context for the LLM. Designing an effective memory module involves several challenges, including how to structure memory representations, when and what to store, how to retrieve relevant information efficiently, and how to integrate it into the reasoning process Zeng et al. (2024a). For a more comprehensive review of memory mechanisms in AI agents, we refer readers to the survey by Zhang et al. (2024d).

(4) Tool Use. The ability to use external tools is a key factor for AI agents to effectively operate in real-world scenarios. While LLMs are powerful in language understanding and generation, their capabilities are inherently limited by their static knowledge and reasoning capabilities. By using external tools, agents can extend their functional scope, allowing them to better interact with real-world environments. Typical tools include web search engines (Li et al., 2025c), code interpreters or execution environments (Islam et al., 2024), and browser automation framework (Müller and Žunič, 2024). The design of the tool-use component often involves selecting tools, constructing tool-specific inputs, invoking API calls, and integrating tool outputs back into the reasoning process.