<?xml version="1.0" encoding="UTF-8" ?>
  <rss version="2.0">
    <channel>
        <title>Xisen Wang</title>
        <link>https://www.wangxisen.com</link>
        <description>Essays and research notes by Xisen Wang</description>
        <item>
          <title>Agent of Agents: The Future of AI Systems</title>
          <link>https://www.wangxisen.com/blogs/agent-of-agents</link>
          <description>Exploring the concept of agent-based AI systems and their potential to revolutionize how we approach complex problems.</description>
          <pubDate>Wed, 20 Nov 2024 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Agentic Graph Engineering: Hold the Agents Fixed, Improve the Graph</title>
          <link>https://www.wangxisen.com/blogs/agentic-graph-engineering</link>
          <description>Now that independently built agents run together on SharedNet, one question sits above the rest: if you hold the agents fixed, what is the best substrate for connecting them? My current answer is to close the gap between Room mode, where people connect agents and set their roles by hand, and Go mode, where a goal would find, connect and prompt them.</description>
          <pubDate>Sat, 03 Oct 2026 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>AI Research Automation: The Next Frontier</title>
          <link>https://www.wangxisen.com/blogs/ai-research-automation</link>
          <description>How AI is transforming scientific research through automated experimentation, analysis, and discovery.</description>
          <pubDate>Thu, 12 Sep 2024 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Do Agent Teams Just Try Harder?</title>
          <link>https://www.wangxisen.com/blogs/do-agent-teams-just-try-harder</link>
          <description>One LLM agent against teams of three and ten, speeding up a minimum-cost flow solver and two more tasks, graded on unseen inputs. Keeping agents working mattered more than adding agents, and ten agents beat three only where the task offered many independent improvements.</description>
          <pubDate>Sat, 24 Oct 2026 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>General Personal Executive Assistant: The New Digital Necessity</title>
          <link>https://www.wangxisen.com/blogs/general-personal-executive-assistant</link>
          <description>Why truly agentic AI assistants are becoming essential tools for modern productivity, transforming from passive responders to proactive partners.</description>
          <pubDate>Fri, 11 Apr 2025 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Fragmented Cognition: Why do we need GPT with Admin Context?</title>
          <link>https://www.wangxisen.com/blogs/gpt-admin-context</link>
          <description>Exploring how AI with privileged access to personal digital ecosystems can transform productivity by defragmenting cognition and creating a unified, proactive cognitive layer.</description>
          <pubDate>Thu, 14 Aug 2025 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Heuristic Learning for Image Classification: A Symbolic Phase 1 Without Neural Networks</title>
          <link>https://www.wangxisen.com/blogs/hl-imagenet-phase1</link>
          <description>Can a coding agent grow a purely symbolic ImageNet classifier from feedback, with no gradients and no neural networks? Phase 1 of HL-ImageNet.</description>
          <pubDate>Tue, 19 May 2026 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Phase 2: Heuristic Learning for Symbolic ImageNet-10</title>
          <link>https://www.wangxisen.com/blogs/hl-imagenet-phase2</link>
          <description>What happened when Claude Code and Codex iteratively improved a non-neural symbolic vision system on a stricter 10-class ImageNet task.</description>
          <pubDate>Mon, 25 May 2026 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>The Loop Is the New Prompt. The Network Is the New Loop.</title>
          <link>https://www.wangxisen.com/blogs/loop-is-the-new-prompt</link>
          <description>Loop engineering made a single agent autonomous over time, but a loop alone stalls the moment it needs something only someone else knows. The next unit of work is a network of permissioned agents, and the memory that makes it compound lives on the edges between them.</description>
          <pubDate>Thu, 09 Jul 2026 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Memory, Context Curation, and the Next Look of Agentic Systems</title>
          <link>https://www.wangxisen.com/blogs/memory-context-agentic-systems</link>
          <description>Why memory and context curation are the new cornerstones of agentic AI systems—and why the future of intelligence depends on how we store, curate, and share knowledge.</description>
          <pubDate>Tue, 15 Apr 2025 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>From Personal COO to the Agent Network</title>
          <link>https://www.wangxisen.com/blogs/personal-coo-to-agent-network</link>
          <description>AI is shifting from answering questions to executing work. The decisive frontier is not isolated agents that perform well in demos, but an operating layer that makes agents dependable under real deployment constraints.</description>
          <pubDate>Sat, 17 Jan 2026 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Personal Manifesto: 2026</title>
          <link>https://www.wangxisen.com/blogs/personal-manifesto-2026</link>
          <description>Intelligence is no longer about building better models. It is about designing systems. The next transformation is the Network of Agents—and it begins with understanding how intelligence itself is organised.</description>
          <pubDate>Sat, 17 Jan 2026 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Keeping Agents Reachable: SharedNet wait, State Management, and Runtime Integration</title>
          <link>https://www.wangxisen.com/blogs/sharednet-wait-state-machines</link>
          <description>A practical guide to SharedNet wait, watch, and serve: message acknowledgments, session ownership, Codex queue delivery, Claude Code resume, and service supervision. Based on a real migration incident, with four diagrams, deployment examples, and a reusable agent instruction.</description>
          <pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Startup Diary 01: On Competitor Analysis</title>
          <link>https://www.wangxisen.com/blogs/startup-diary-01-competitor-analysis</link>
          <description>Why having competitors is a blessing, not a curse, and how deep competitor analysis shapes startup strategy.</description>
          <pubDate>Wed, 16 Apr 2025 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Systems Intelligence: The Future of AI Beyond Single Models</title>
          <link>https://www.wangxisen.com/blogs/systems-intelligence</link>
          <description>Why interconnected, self-evolving systems represent the next frontier of artificial intelligence, surpassing the limitations of isolated model development.</description>
          <pubDate>Thu, 10 Apr 2025 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Welcome to My Digital Space</title>
          <link>https://www.wangxisen.com/blogs/welcome-to-my-digital-space</link>
          <description>A short note on who I am and what I write about here.</description>
          <pubDate>Mon, 20 May 2024 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>WeLight: Illuminating the Path to Educational Excellence</title>
          <link>https://www.wangxisen.com/blogs/welight-journey</link>
          <description>The journey of building WeLight from an Oxford dorm room idea to an AI-powered education platform transforming how students navigate their academic futures.</description>
          <pubDate>Sat, 15 Feb 2025 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Organisational Intelligence: Why I Started Systemind</title>
          <link>https://www.wangxisen.com/blogs/why-i-started-systemind</link>
          <description>Making agents smarter does not make them better at working together. A short note on the question I keep coming back to, and why I started a project to test it.</description>
          <pubDate>Sun, 20 Sep 2026 00:00:00 GMT</pubDate>
        </item>
<item>
          <title>Why Does a Simple Abstraction Matter?</title>
          <link>https://www.wangxisen.com/blogs/why-simple-abstraction-matters</link>
          <description>All of 2026 has been a year of subtraction. Cutting 99% of our product taught me that subtraction is a harder version of addition, and that a simple, general abstraction is the most powerful thing a founder, a researcher, or an infrastructure builder can ship.</description>
          <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>
        </item>
    </channel>
  </rss>