作者 | 周星星 编辑 | 大模型之心Tech 原文链接： https://zhuanlan.zhihu.com/p/1983512173549483912 点击下方 卡片 ，关注" 大模型之心Tech "公众号 戳我-> 领取大模型巨卷干货 本文只做学术分享，已获转载授权 ，欢迎添加小助理微信AIDriver004做进一步咨询 前言 随着 2025 年即将画上句号，我想对"Agent 元年"根据个人这一年的实践和认知进行一次收敛。 技术观点：Agent 架构之争已定，收敛至以 Claude Code 和 Deep Agent 为代表的「通用型 Agent」形态。 Claude Code 虽然在 2025 年 3 月作为"智能终端编程助手"推出，但其不止于编程。 行业认知: 2025 年作为 Agent 元年，既没有悲观者眼中的"名不副实"，也未完全达到乐观者预期的"全面替代"，而是处于稳步落地的中间态。 作为一线从业者，我的评价是： 技术已就绪，爆发在局部 。 基于以上背景，本文将从 Deep Agent 为切入点，分享我作为一线开发者在 2025 年的实战感悟。 主要参考资料： Anthropic、Lan ...

Core Insights - The article discusses the evolution of memory systems in AI agents, emphasizing the transition from optional modules to essential infrastructure for various applications such as conversational assistants and code engineering [2] - A comprehensive survey titled "Memory in the Age of AI Agents: A Survey" has been published by leading academic institutions to provide a unified perspective on the rapidly expanding yet fragmented concept of "Agent Memory" [2] Forms of Memory - The survey categorizes agent memory into three main forms: token-level, parametric, and latent memory, focusing on how information is represented, stored, and accessed [16][24] - Token-level memory is defined as persistent, discrete units that are externally accessible and modifiable, making it the most explicit form of memory [18] - Parametric memory involves storing information within model parameters, which can lead to challenges in retrieval and updating due to its flat structure [22] - Latent memory exists in the model's internal states and can be continuously updated during inference, providing a compact representation of memory [24][26] Functions of Memory - The article identifies three core functions of agent memory: factual memory, experiential memory, and working memory [29] - Factual memory aims to provide a stable reference for knowledge acquired from user interactions and environmental facts, ensuring consistency across sessions [31] - Experiential memory focuses on accumulating knowledge from past interactions to enhance problem-solving capabilities, allowing agents to learn from experiences [32] - Working memory manages information within single task instances, addressing the challenge of processing large and complex inputs [35] Dynamics of Memory - The dynamics of memory encompass three stages: formation, evolution, and retrieval, which form a feedback loop [38] - The formation stage encodes raw context into more compact knowledge representations, addressing computational and memory constraints [40] - The evolution stage integrates new memories with existing ones, ensuring coherence and efficiency through mechanisms like pruning and conflict resolution [43] - The retrieval stage determines how memory can assist in decision-making, emphasizing the importance of effective querying strategies [41] Future Directions - The article suggests that future memory systems should be viewed as a core capability of agents rather than mere retrieval plugins, integrating memory management into decision-making processes [49][56] - There is a growing emphasis on automating memory management, allowing agents to self-manage their memory operations, which could lead to more robust and adaptable systems [54][62] - The integration of reinforcement learning into memory control is highlighted as a potential pathway for developing more sophisticated memory systems that can learn and optimize over time [58][60]

Hi everyone, welcome back to another build hour. I'm Michaela on the startup marketing team and I'm here today with two members of our solution architecture team. Emry live in the studio and Brian joining virtually to help address Q&A throughout the hour. >> Hi, I'm Emry. I work as a solution architect at OpenAI supporting digital native customers on building various of AI use cases including longunning AI agents. So today's topic is agent memory patterns which is a very exciting topic in Emry and I's first ...

Hey folks, it's Sydney from Linkchain and I'm super excited to be back with another episode in our Python middleware series. This time we're going to be covering our summarization middleware. So context engineering is all the rage these days.But what does that actually mean. Context engineering is giving your model, which powers your agent, the right information and tools at the right time so that it can execute a given task. One of the most important tools that you can use to optimize your context for an a ...

Hi there, this is Christian from Lchain. If you build with coding agents like cursor, you probably recognize this. The first few turns with the agents are great.But then as you keep continuing talking to the agent in the same thread, the quality slides, the decision get more fuzzy and the overall code quality drops and then cursor drops this system line context summarized. That's the moment you know you've crossed the context boundary line. So why is summarization such a big deal for context engineering.Eve ...

To summarise, I'll leave you with the context engineering graphic.Also, here's an open-source stack for context engineering:- Memory: @zep_ai- Knowledge base: @milvusio- Agent orchestration: @crewAIInc- Observability & tracing: @deepeval https://t.co/qseVsjlXBr ...

Context engineering, clearly explained (with visuals):(an illustrated guide below) https://t.co/am97ee4vTA ...

Right. Hello everyone. Uh today I will present meta adaptive context engineering or meta AC for short which is a new framework designed to optimize AI agents beyond single dimension approaches.We will explore how orchestrating multiple adaptation strategies can overcome the limitations of existing context engineering methods. Now a little introduction about myself. Uh so I'm Alberto Romero.I'm the co-founder and CEO at jointly. And for context at jointly we build the main specialized agents for regulated in ...

Context Engineering & AI - Context engineering is evolving from simple prompt engineering to a dynamic approach that feeds AI with wider context for better results [3] - Context engineering enables selective curation of information relevant to specific domains, especially important in enterprise environments [4] - Structuring input in context engineering improves signal over noise, addressing a major problem with current AI models [5] - Memory, both short-term and long-term, is crucial for AI, enabling collaboration, remembering conversation history, and effective long-term operations [10][11][12] Knowledge Graphs & Graph RAG - Knowledge graphs provide structured information that complements AI's ability to create and pull from different sources [17] - Graph RAG, which uses graphs as part of the retrieval process, provides more relevant results than vector similarity search by incorporating relationships, nodes, and community groupings [22][23] - Graph RAG enables explainable AI and allows for the implementation of role-based access control, ensuring that only authorized individuals can access specific information [25] Neo4j Solutions & Resources - Neo4j offers a knowledge graph builder, a web application that allows users to upload files and generate knowledge graphs [28] - Neo4j's MCP server is an open-source extension that enables querying knowledge graphs using Cypher, a graph query language [46] - Neo4j provides resources like Graph Academy (free learning resources) and Nodes AI (virtual conference) for learning about graph technology and AI applications [53][54]

Core Insights - Google’s Gemini 3 has transformed the AI landscape in Silicon Valley, positioning the company as a leader rather than a follower in the AI race against OpenAI and Anthropic [1][3] - Gemini 3 is recognized for its significant advancements in multimodal capabilities and is seen as a prime example of executing Scaling Law effectively [1][3] Performance Evaluation - Within 48 hours of its release, Gemini 3 topped various performance rankings, showcasing its true multimodal native model capabilities [4][6] - Users reported that Gemini 3 provides a more integrated development experience, particularly with tools like Google AntiGravity, which enhances coding efficiency by allowing simultaneous visual and coding tasks [6][7] Technical Innovations - The model achieved a notable improvement in Few-shot Learning, reaching over 30% on the ARC-AGI-2 Benchmark, indicating a qualitative leap in its reasoning capabilities [10][11] - Gemini 3 employs a tree-based thought process and self-rewarding mechanisms, allowing it to explore multiple reasoning paths simultaneously [19][20] Developer Ecosystem - The release of Gemini 3 and AntiGravity has led to discussions about the end of the coding competition, as Google’s ecosystem may create significant barriers for startups like Cursor [22][23] - Despite the strong capabilities of AntiGravity, it still faces challenges in backend deployment and complex system architecture, suggesting that independent developers may still find opportunities in niche areas [25][26] Future Trends in AI - The focus is shifting towards new AI paradigms beyond LLMs, with emerging labs like NeoLab attracting significant venture capital [27][28] - There is a growing interest in developing world models that understand physical laws, indicating a potential shift in AI research directions [31][32] Conclusion - The launch of Gemini 3 serves as a robust counter to the "AI bubble" narrative, demonstrating that with sufficient computational power and engineering optimization, Scaling Law remains a viable path for AI advancement [32][33]