<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Ken Lin on Medium]]></title>
        <description><![CDATA[Stories by Ken Lin on Medium]]></description>
        <link>https://medium.com/@ken_lin?source=rss-97ef628d7e1a------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*4Qd-d7RuQTGvQpNYw-L4Zg.jpeg</url>
            <title>Stories by Ken Lin on Medium</title>
            <link>https://medium.com/@ken_lin?source=rss-97ef628d7e1a------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 18:18:54 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ken_lin/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Microsoft Prompt Flow: The All-in-One Tool for Streamlining LLM Workflows]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://aiadvances.org/microsoft-prompt-flow-the-all-in-one-tool-for-streamlining-llm-workflows-020abe927b27?source=rss-97ef628d7e1a------2"><img src="https://cdn-images-1.medium.com/max/1507/1*IrLfZ_62zqTqPZp3Db6b3A.png" width="1507"></a></p><p class="medium-feed-snippet">Get the clarity you need with visual workflows that lay out your entire development process.</p><p class="medium-feed-link"><a href="https://aiadvances.org/microsoft-prompt-flow-the-all-in-one-tool-for-streamlining-llm-workflows-020abe927b27?source=rss-97ef628d7e1a------2">Continue reading on AI Advances »</a></p></div>]]></description>
            <link>https://aiadvances.org/microsoft-prompt-flow-the-all-in-one-tool-for-streamlining-llm-workflows-020abe927b27?source=rss-97ef628d7e1a------2</link>
            <guid isPermaLink="false">https://medium.com/p/020abe927b27</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[promptflow]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Ken Lin]]></dc:creator>
            <pubDate>Mon, 31 Mar 2025 11:04:22 GMT</pubDate>
            <atom:updated>2025-03-31T11:04:22.737Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Python: Fast to Develop, Slow to Run]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@ken_lin/python-fast-to-develop-slow-to-run-62144903d735?source=rss-97ef628d7e1a------2"><img src="https://cdn-images-1.medium.com/max/1024/1*W-bQxQIcP1PN5tjVeEsPcw.jpeg" width="1024"></a></p><p class="medium-feed-snippet">Are you trading execution speed for coding speed?</p><p class="medium-feed-link"><a href="https://medium.com/@ken_lin/python-fast-to-develop-slow-to-run-62144903d735?source=rss-97ef628d7e1a------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@ken_lin/python-fast-to-develop-slow-to-run-62144903d735?source=rss-97ef628d7e1a------2</link>
            <guid isPermaLink="false">https://medium.com/p/62144903d735</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Ken Lin]]></dc:creator>
            <pubDate>Thu, 20 Mar 2025 15:17:07 GMT</pubDate>
            <atom:updated>2025-03-20T15:17:07.168Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Forget Traditional Web Scrapers — Crawl4AI is Engineered for AI, LLMs, and Data Processing]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://generativeai.pub/forget-traditional-web-scrapers-crawl4ai-is-engineered-for-ai-llms-and-data-processing-f1cb39e9d29e?source=rss-97ef628d7e1a------2"><img src="https://cdn-images-1.medium.com/max/1000/1*ewM5XqxdaZcvOhBy3hf3hw.jpeg" width="1000"></a></p><p class="medium-feed-snippet">Smarter Crawling Leads to More Efficient Data Collection and LLM-Optimized Data Formats</p><p class="medium-feed-link"><a href="https://generativeai.pub/forget-traditional-web-scrapers-crawl4ai-is-engineered-for-ai-llms-and-data-processing-f1cb39e9d29e?source=rss-97ef628d7e1a------2">Continue reading on Generative AI »</a></p></div>]]></description>
            <link>https://generativeai.pub/forget-traditional-web-scrapers-crawl4ai-is-engineered-for-ai-llms-and-data-processing-f1cb39e9d29e?source=rss-97ef628d7e1a------2</link>
            <guid isPermaLink="false">https://medium.com/p/f1cb39e9d29e</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[crawl4ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Ken Lin]]></dc:creator>
            <pubDate>Wed, 19 Mar 2025 16:17:35 GMT</pubDate>
            <atom:updated>2025-03-29T13:40:27.932Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Browser-Use: AI Web Browsing Is Here — But It’s Far From Perfect]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@ken_lin/browser-use-ai-web-browsing-is-here-but-its-far-from-perfect-426aa8720036?source=rss-97ef628d7e1a------2"><img src="https://cdn-images-1.medium.com/max/1024/1*Dwu9dYO6mURkKVCphifrHg.jpeg" width="1024"></a></p><p class="medium-feed-snippet">Browser-use is a Python package designed to seamlessly connect AI agents with browser functionality, enabling automated web interactions&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@ken_lin/browser-use-ai-web-browsing-is-here-but-its-far-from-perfect-426aa8720036?source=rss-97ef628d7e1a------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@ken_lin/browser-use-ai-web-browsing-is-here-but-its-far-from-perfect-426aa8720036?source=rss-97ef628d7e1a------2</link>
            <guid isPermaLink="false">https://medium.com/p/426aa8720036</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[programming]]></category>
            <dc:creator><![CDATA[Ken Lin]]></dc:creator>
            <pubDate>Mon, 17 Mar 2025 14:57:42 GMT</pubDate>
            <atom:updated>2025-03-18T13:57:40.374Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[LangGraph vs. LangChain: The Evolution of AI Workflow Orchestration]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@ken_lin/langgraph-vs-langchain-the-evolution-of-ai-workflow-orchestration-78d05a8441cb?source=rss-97ef628d7e1a------2"><img src="https://cdn-images-1.medium.com/max/1024/1*6t-TAEJ9byALt4aT3RRkaw.jpeg" width="1024"></a></p><p class="medium-feed-snippet">LangGraph has emerged as a powerful extension of the LangChain ecosystem, offering capabilities that significantly surpass those of its&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@ken_lin/langgraph-vs-langchain-the-evolution-of-ai-workflow-orchestration-78d05a8441cb?source=rss-97ef628d7e1a------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@ken_lin/langgraph-vs-langchain-the-evolution-of-ai-workflow-orchestration-78d05a8441cb?source=rss-97ef628d7e1a------2</link>
            <guid isPermaLink="false">https://medium.com/p/78d05a8441cb</guid>
            <category><![CDATA[llm-applications]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[langchain]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[langgraph]]></category>
            <dc:creator><![CDATA[Ken Lin]]></dc:creator>
            <pubDate>Sun, 16 Mar 2025 04:37:06 GMT</pubDate>
            <atom:updated>2025-03-18T13:42:10.696Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Why Smart Developers Are Moving Away from LangChain]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@ken_lin/why-smart-developers-are-moving-away-from-langchain-9ee97d988741?source=rss-97ef628d7e1a------2"><img src="https://cdn-images-1.medium.com/max/1024/1*xIPKy3dh16fwovyq-4vMPw.jpeg" width="1024"></a></p><p class="medium-feed-snippet">LangChain has emerged as one of the most widely discussed frameworks for building applications with large language models (LLMs), but&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@ken_lin/why-smart-developers-are-moving-away-from-langchain-9ee97d988741?source=rss-97ef628d7e1a------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@ken_lin/why-smart-developers-are-moving-away-from-langchain-9ee97d988741?source=rss-97ef628d7e1a------2</link>
            <guid isPermaLink="false">https://medium.com/p/9ee97d988741</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[langchain]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[llm-applications]]></category>
            <dc:creator><![CDATA[Ken Lin]]></dc:creator>
            <pubDate>Sat, 15 Mar 2025 16:03:11 GMT</pubDate>
            <atom:updated>2025-03-18T13:44:49.157Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Introduction to Microsoft GraphRAG]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@ken_lin/microsoft-graphrag-b7ec58acdd83?source=rss-97ef628d7e1a------2"><img src="https://cdn-images-1.medium.com/max/1024/1*UPwqI7lChxQMKKTGIxydRg.jpeg" width="1024"></a></p><p class="medium-feed-snippet">Microsoft GraphRAG (Graphs + Retrieval-Augmented Generation) is an advanced framework that enhances traditional RAG by leveraging&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@ken_lin/microsoft-graphrag-b7ec58acdd83?source=rss-97ef628d7e1a------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@ken_lin/microsoft-graphrag-b7ec58acdd83?source=rss-97ef628d7e1a------2</link>
            <guid isPermaLink="false">https://medium.com/p/b7ec58acdd83</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm-applications]]></category>
            <category><![CDATA[graphrag]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[knowledge-graph]]></category>
            <dc:creator><![CDATA[Ken Lin]]></dc:creator>
            <pubDate>Fri, 14 Mar 2025 15:46:01 GMT</pubDate>
            <atom:updated>2025-03-18T09:27:08.433Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[LangGraph: A Framework for Building Stateful Multi-Agent LLM Applications]]></title>
            <link>https://medium.com/@ken_lin/langgraph-a-framework-for-building-stateful-multi-agent-llm-applications-a51d5eb68d03?source=rss-97ef628d7e1a------2</link>
            <guid isPermaLink="false">https://medium.com/p/a51d5eb68d03</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[langgraph]]></category>
            <category><![CDATA[pipeline]]></category>
            <category><![CDATA[llm-applications]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Ken Lin]]></dc:creator>
            <pubDate>Thu, 13 Mar 2025 15:06:46 GMT</pubDate>
            <atom:updated>2025-03-18T09:28:52.131Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/988/1*MEfWmO_J-LXJaSUk1K3BWw.jpeg" /><figcaption>Image generated by OpenAI DALL·E</figcaption></figure><blockquote>LangGraph is a powerful Python library designed for constructing stateful, multi-actor applications with Large Language Models (LLMs). It extends the capabilities of LangChain Expression Language (LCEL) while specifically addressing limitations in existing frameworks for agent development. As a specialized tool for creating complex LLM applications, LangGraph provides a structured approach to building sophisticated workflows that require cyclical processing patterns.</blockquote><h3><strong>Fundamental Design Principles</strong></h3><p>LangGraph is built on top of and intended to be used with LangChain, extending the LangChain Expression Language with capabilities to coordinate multiple chains or actors across computational steps in a cyclic manner. The library is inspired by graph processing frameworks like Pregel and Apache Beam, offering an interface similar to NetworkX for building complex agent systems. Unlike purely Directed Acyclic Graph (DAG) frameworks, LangGraph’s main distinctive feature is its support for cycles, which are essential for building agent-like behaviors where LLMs need to be called in loops to determine subsequent actions.</p><p>The architecture of LangGraph fundamentally revolves around modeling applications as directed graphs. Each node in the graph represents an LLM agent or computational component, while edges represent the communication channels or information flow between these components. This structured approach enables developers to create clear, manageable workflows where components perform specific tasks and pass information to other components as needed, supporting both linear sequences and complex conditional branching.</p><h3><strong>State Management</strong></h3><p>One of LangGraph’s core strengths is its automatic state management capability, which allows tracking and persistence of information across multiple interactions within an application. As agents perform tasks, the state is dynamically updated, ensuring the system maintains context and responds appropriately to new inputs. This feature is particularly valuable in multi-step agent workflows where context from previous steps influences future decisions, creating a more coherent and continuous experience.</p><p>The state management in LangGraph handles the complexity of maintaining conversation history, tool outputs, and intermediate reasoning steps without requiring developers to implement custom state tracking mechanisms. This reduces cognitive load when building complex applications and helps prevent state-related bugs that might otherwise occur in multi-agent systems.</p><h3><strong>Setting Up a LangGraph System</strong></h3><p>Creating a functional LangGraph application begins with installation through pip: pip install langgraph. The implementation process involves defining nodes, establishing connections between them, and specifying the execution flow. Nodes in LangGraph can represent various components such as LLM agents, tool executors, or data processors, each responsible for specific functionality within the overall system.</p><p>The Graph class serves as the primary interface for constructing LangGraph applications. Through methods like add_node, add_edge, and add_conditional_edges, developers can define the structure and behavior of their agent workflows. The set_entry_point method designates the starting node for the graph execution, while the compile method transforms the graph definition into a runnable LangChain object.</p><h3><strong>Defining Workflow Logic</strong></h3><p>A typical LangGraph workflow definition includes several key components. First, nodes are added to represent the main actors or processing steps in the application. Then, edges are established to define the flow of information between nodes. Conditional edges allow for dynamic routing based on the output of nodes, enabling adaptive behaviors that can respond appropriately to different situations.</p><p>For example, in an agent workflow, one might define an “agent” node that decides what action to take and a “tools” node that executes those actions. A conditional edge from the agent node could either continue to the tools node or exit the workflow based on the agent’s decision, with a regular edge returning from the tools back to the agent to form a cycle. This cyclical structure enables the iterative decision-making characteristic of agent systems.</p><h3><strong>Cyclical Workflow Support</strong></h3><p>The most significant advantage of LangGraph is its explicit support for cyclical workflows, which are essential for implementing agent-like behaviors. As emphasized in the documentation: “Crucially, this is NOT a DAG framework… Cycles are important for agent-like behaviors, where you call an LLM in a loop, asking it what action to take next.” This capability enables the development of sophisticated agents that can iteratively refine their understanding and actions, leading to more capable AI systems.</p><p>Traditional DAG frameworks struggle with these types of applications because they fundamentally cannot represent the looping patterns needed for agent decision-making. LangGraph bridges this gap, making it substantially easier to implement complex agent behaviors that require multiple rounds of thinking, tool use, and refinement.</p><h3><strong>Streamlined Multi-Agent Coordination</strong></h3><p>LangGraph provides a clear and intuitive mechanism for defining coordination between multiple agents or components in a system. The graph-based structure visually represents how information flows between components, making it easier to design, implement, and reason about complex multi-agent workflows. This explicit coordination model helps prevent common issues in multi-agent systems such as race conditions, deadlocks, or communication failures.</p><p>The framework’s state management capabilities further enhance multi-agent coordination by ensuring that all agents have access to the necessary context and can build upon each other’s work. This creates cohesive systems where specialized agents can collaborate effectively to solve complex problems.</p><h3><strong>Advanced Streaming Capabilities</strong></h3><p>LangGraph includes built-in support for streaming output from nodes as they are produced. This streaming capability allows for real-time monitoring of workflow execution, providing visibility into the intermediate states and decisions made by each component. Developers can observe how information flows through the system and how agents respond to different inputs, which is invaluable for debugging and understanding complex agent behaviors.</p><p>The streaming functionality can be leveraged for user interfaces that display intermediate results or thought processes, creating more transparent AI systems where users can follow the reasoning path that led to particular conclusions or recommendations.</p><h3><strong>Increased Complexity for Simple Applications</strong></h3><p>For straightforward applications that don’t require cyclical workflows or multi-agent coordination, LangGraph may introduce unnecessary complexity. The overhead of defining graphs, managing state, and coordinating components might not be justified for simple use cases where a linear chain of operations would suffice. Developers should consider whether their application truly needs the advanced features LangGraph provides or if a simpler approach would be more appropriate.</p><p>The additional abstraction layer introduced by the graph-based programming model can make simple logic more difficult to follow compared to procedural code or basic chain structures. This complexity becomes worthwhile in sophisticated applications but could be a disadvantage in more straightforward scenarios.</p><h3><strong>Learning Curve and Development Overhead</strong></h3><p>The graph-based programming paradigm, while powerful, presents a learning curve for developers new to this approach or to the LangChain ecosystem. Understanding how to effectively structure graphs, manage state, and implement conditional logic within the LangGraph framework requires time and experience. Organizations adopting LangGraph may need to invest in training and documentation to bring their development teams up to speed.</p><p>Additionally, the framework’s abstraction can sometimes make debugging more challenging, as issues may arise from the interaction between components rather than within individual components. This distributed nature of problems can increase the effort required to identify and resolve issues in complex graphs.</p><h3><strong>Integration Challenges</strong></h3><p>While LangGraph integrates well with the LangChain ecosystem, incorporating it into existing systems or workflows may require significant adaptation. The framework’s assumptions about state management and execution flow might not align perfectly with other frameworks or custom implementations, necessitating additional integration work.</p><p>For organizations with established AI infrastructure, adopting LangGraph might involve refactoring existing code or creating adapter layers to bridge between different paradigms. This integration effort should be factored into decisions about whether to adopt LangGraph for specific projects.</p><h3><strong>Complex Conversational Agents</strong></h3><p>LangGraph excels in building sophisticated conversational agents that maintain context, make decisions, and adapt their behavior based on user interactions. The cyclical nature of conversation, where each exchange builds on previous context, aligns perfectly with LangGraph’s support for stateful, iterative workflows. These agents can engage in multi-turn conversations, use tools to gather information, and progressively refine their responses based on user feedback.</p><p>The framework’s state management capabilities ensure that conversational context is maintained across interactions, while its conditional logic allows agents to handle diverse user inputs appropriately. This makes LangGraph particularly valuable for building virtual assistants, customer service bots, or educational agents that need to engage in extended, meaningful dialogues.</p><h3><strong>Research and Simulation Environments</strong></h3><p>Multi-agent systems built with LangGraph can be used to simulate complex environments where different actors interact according to specific rules or objectives. These simulations can model economic systems, social dynamics, or competitive scenarios, providing insights into emergent behaviors and complex interactions.</p><p>The explicit representation of agent relationships and information flow in LangGraph makes it well-suited for research applications where understanding and visualizing system dynamics is important. Researchers can analyze how different agent configurations and communication patterns affect overall system behavior, leading to insights that might not be apparent from studying individual agents in isolation.</p><h3><strong>Decision Support Systems</strong></h3><p>LangGraph’s support for cyclical workflows makes it ideal for building decision support systems that involve iterative reasoning, information gathering, and analysis. These systems can help professionals in fields like medicine, finance, or law navigate complex decisions by breaking problems down into manageable steps, gathering relevant information through tools, and synthesizing insights from multiple sources.</p><p>The framework’s ability to model complex reasoning processes as explicit graphs allows for more transparent decision-making, where users can understand the steps and considerations that led to particular recommendations. This transparency is particularly valuable in high-stakes domains where explainability is a key requirement.</p><h3><strong>Conclusion</strong></h3><p>LangGraph represents a significant advancement in the toolkit available for developing sophisticated LLM applications, particularly those involving agent-like behaviors and multi-agent coordination. By extending LangChain with support for cyclical workflows and robust state management, it addresses key limitations of earlier approaches and enables new possibilities for AI systems that can reason, act, and interact in more complex ways.</p><p>While the framework introduces some complexity and requires investment to master, its benefits for developing advanced AI applications make it a valuable addition to the LangChain ecosystem. As LLM applications continue to grow in sophistication, tools like LangGraph will play an increasingly important role in helping developers create more capable, flexible, and reliable AI systems.</p><p>For organizations and developers looking to build complex agent systems or applications requiring stateful, iterative processing, LangGraph provides a robust foundation that can significantly reduce development time and improve system quality. However, those working on simpler applications should carefully evaluate whether the additional complexity is justified by their specific requirements and use cases.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a51d5eb68d03" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>