<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Agent on Leafw&#39;s Blog</title>
    <link>https://leafw.net/en/tags/agent/</link>
    <description>Recent content in Agent on Leafw&#39;s Blog</description>
    <generator>Hugo</generator>
    <language>en</language>
    <managingEditor>wyr95626@gmail.com (Leafw)</managingEditor>
    <webMaster>wyr95626@gmail.com (Leafw)</webMaster>
    <copyright>This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.</copyright>
    <lastBuildDate>Sun, 08 Nov 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://leafw.net/en/tags/agent/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Anthropic Taught Agents to &#39;Dream&#39;: Rethinking Long-Term Memory for AI</title>
      <link>https://leafw.net/en/anthropic-agent-memory/</link>
      <pubDate>Sun, 08 Nov 2026 00:00:00 +0000</pubDate><author>wyr95626@gmail.com (Leafw)</author>
      <guid>https://leafw.net/en/anthropic-agent-memory/</guid>
      <description>&lt;p&gt;&lt;img&#xA;        class=&#34;lazyload&#34;&#xA;        src=&#34;https://leafw.net/svg/loading.min.svg&#34;&#xA;        data-src=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/anthropic-agent-memory-cover.png&#34;&#xA;        data-srcset=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/anthropic-agent-memory-cover.png, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/anthropic-agent-memory-cover.png 1.5x, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/anthropic-agent-memory-cover.png 2x&#34;&#xA;        data-sizes=&#34;auto&#34;&#xA;        alt=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/anthropic-agent-memory-cover.png&#34;&#xA;        title=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/anthropic-agent-memory-cover.png&#34; /&gt;&lt;/p&gt;&#xA;&lt;p&gt;On November 6, 2026, Anthropic published a technical article titled &lt;a href=&#34;https://www.anthropic.com/research/components-of-agent-memory&#34; target=&#34;_blank&#34; rel=&#34;noopener noreffer &#34;&gt;&lt;em&gt;The components of agent memory&lt;/em&gt;&lt;/a&gt;. It tackles a question that is becoming increasingly important in agent development: &lt;strong&gt;how do you give an AI agent long-term memory that is actually valuable?&lt;/strong&gt;&lt;/p&gt;&#xA;&lt;p&gt;When people hear &amp;ldquo;agent memory,&amp;rdquo; the first thing that comes to mind is usually saving conversation history to a database, then using embeddings and vector search to pull relevant pieces back into the context when needed.&lt;/p&gt;</description>
    </item>
    <item>
      <title>OpenAI Is Rebuilding Git, Too</title>
      <link>https://leafw.net/en/openai-pando/</link>
      <pubDate>Tue, 03 Nov 2026 00:00:00 +0000</pubDate><author>wyr95626@gmail.com (Leafw)</author>
      <guid>https://leafw.net/en/openai-pando/</guid>
      <description>&lt;p&gt;&lt;img&#xA;        class=&#34;lazyload&#34;&#xA;        src=&#34;https://leafw.net/svg/loading.min.svg&#34;&#xA;        data-src=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/openai-pando-cover.png&#34;&#xA;        data-srcset=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/openai-pando-cover.png, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/openai-pando-cover.png 1.5x, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/openai-pando-cover.png 2x&#34;&#xA;        data-sizes=&#34;auto&#34;&#xA;        alt=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/openai-pando-cover.png&#34;&#xA;        title=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/openai-pando-cover.png&#34; /&gt;&lt;/p&gt;&#xA;&lt;p&gt;On October 13, Anthropic published &lt;a href=&#34;https://www.anthropic.com/engineering/redesigning-source-control&#34; target=&#34;_blank&#34; rel=&#34;noopener noreffer &#34;&gt;Redesigning source control at Anthropic&lt;/a&gt;, introducing Scion, its in-house version control system. I wrote about it at the time in &lt;a href=&#34;https://leafw.net/en/anthropic-scion/&#34; rel=&#34;&#34;&gt;Anthropic&amp;rsquo;s In-House Version Control: Scion&lt;/a&gt;. The core issue was that Claude produces code so fast that Git, builds, and CI, infrastructure originally built for human developers, had started to become bottlenecks.&lt;/p&gt;&#xA;&lt;p&gt;Less than a month later, on November 2, OpenAI published its own technical post, &lt;a href=&#34;https://openai.com/index/scaling-git-at-openai/&#34; target=&#34;_blank&#34; rel=&#34;noopener noreffer &#34;&gt;Scaling Git at OpenAI&lt;/a&gt;, introducing Pando, the Git backend they built.&lt;/p&gt;</description>
    </item>
    <item>
      <title>DeepSeek Super-agent: Moving Agents from One-off Conversations to Continuous Autonomous Operation</title>
      <link>https://leafw.net/en/deepseek-superagent/</link>
      <pubDate>Sun, 01 Nov 2026 00:00:00 +0000</pubDate><author>wyr95626@gmail.com (Leafw)</author>
      <guid>https://leafw.net/en/deepseek-superagent/</guid>
      <description>&lt;p&gt;&lt;img&#xA;        class=&#34;lazyload&#34;&#xA;        src=&#34;https://leafw.net/svg/loading.min.svg&#34;&#xA;        data-src=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/deepseek-superagent-cover.png&#34;&#xA;        data-srcset=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/deepseek-superagent-cover.png, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/deepseek-superagent-cover.png 1.5x, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/deepseek-superagent-cover.png 2x&#34;&#xA;        data-sizes=&#34;auto&#34;&#xA;        alt=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/deepseek-superagent-cover.png&#34;&#xA;        title=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/deepseek-superagent-cover.png&#34; /&gt;&lt;/p&gt;&#xA;&lt;p&gt;On October 31, the DeepSeek team published an article, &amp;ldquo;Exploring Super-agent Design: Toward More Autonomous Agent Systems,&amp;rdquo; discussing what architecture the next generation of agent systems should adopt.&lt;/p&gt;&#xA;&lt;p&gt;The article also released an open-source research prototype: dsh-superagent-prototype.&lt;/p&gt;&#xA;&lt;p&gt;What makes this article interesting is that it doesn&amp;rsquo;t keep talking about how to improve a model&amp;rsquo;s reasoning, how to design prompts, or how to get multiple agents to collaborate better on a single task. Instead, it focuses on a more fundamental question:&lt;/p&gt;</description>
    </item>
    <item>
      <title>Anthropic: Don&#39;t Rush to Split Out Sub-Agents</title>
      <link>https://leafw.net/en/fold-subagents/</link>
      <pubDate>Sun, 25 Oct 2026 00:00:00 +0000</pubDate><author>wyr95626@gmail.com (Leafw)</author>
      <guid>https://leafw.net/en/fold-subagents/</guid>
      <description>&lt;p&gt;&lt;img&#xA;        class=&#34;lazyload&#34;&#xA;        src=&#34;https://leafw.net/svg/loading.min.svg&#34;&#xA;        data-src=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/fold-subagents-single-loop-cover.png&#34;&#xA;        data-srcset=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/fold-subagents-single-loop-cover.png, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/fold-subagents-single-loop-cover.png 1.5x, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/fold-subagents-single-loop-cover.png 2x&#34;&#xA;        data-sizes=&#34;auto&#34;&#xA;        alt=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/fold-subagents-single-loop-cover.png&#34;&#xA;        title=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/fold-subagents-single-loop-cover.png&#34; /&gt;&lt;/p&gt;&#xA;&lt;p&gt;A few days ago I wrote &lt;a href=&#34;https://www.leafw.net/en/parallel-subagents/&#34; target=&#34;_blank&#34; rel=&#34;noopener noreffer &#34;&gt;a breakdown of a paper on parallel sub-agents&lt;/a&gt;. The takeaway: turning on concurrency generally hurt scores on short tasks, while on long tasks the results were mixed. On October 22, Anthropic published an article on the Claude website, &amp;ldquo;Building agents: When to fold subagents into your main agent loop&amp;rdquo;, written by Anthropic&amp;rsquo;s Diane Sarkis. It&amp;rsquo;s about when you should fold sub-agents back into the main loop, that is, merge a sub-agent&amp;rsquo;s prompt and capabilities into the main agent so they become its skills and tools.&lt;/p&gt;</description>
    </item>
    <item>
      <title>Do More Sub-Agents Make Better Code?</title>
      <link>https://leafw.net/en/parallel-subagents/</link>
      <pubDate>Tue, 20 Oct 2026 00:00:00 +0000</pubDate><author>wyr95626@gmail.com (Leafw)</author>
      <guid>https://leafw.net/en/parallel-subagents/</guid>
      <description>&lt;p&gt;&lt;img&#xA;        class=&#34;lazyload&#34;&#xA;        src=&#34;https://leafw.net/svg/loading.min.svg&#34;&#xA;        data-src=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/parallel-subagents-main-agent-cover.png&#34;&#xA;        data-srcset=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/parallel-subagents-main-agent-cover.png, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/parallel-subagents-main-agent-cover.png 1.5x, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/parallel-subagents-main-agent-cover.png 2x&#34;&#xA;        data-sizes=&#34;auto&#34;&#xA;        alt=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/parallel-subagents-main-agent-cover.png&#34;&#xA;        title=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/parallel-subagents-main-agent-cover.png&#34; /&gt;&lt;/p&gt;&#xA;&lt;p&gt;Many agent products now let a main agent spin up a batch of sub-agents to work in parallel. But how well does multi-agent collaboration actually work, and is it always better than a single agent? Few people bother to compare.&lt;/p&gt;&#xA;&lt;p&gt;I recently read a paper that does exactly that: &lt;em&gt;When Sub-Agents Work in Parallel: The Promises and Pitfalls of Dynamic Concurrency in Long-Horizon Coding Tasks&lt;/em&gt;, &lt;a href=&#34;https://arxiv.org/abs/2610.10263&#34; target=&#34;_blank&#34; rel=&#34;noopener noreffer &#34;&gt;https://arxiv.org/abs/2610.10263&lt;/a&gt; . The work comes from Nanjing University, Central University of Finance and Economics, and the University of Illinois Urbana-Champaign, by Han Li, HanHaoNing Li, Ziqian Jiang, and Yiling Lou. They turned the built-in sub-agent concurrency of three coding agents, Codex, Claude Code, and Kimi Code, on and off, and ran each setting on 354 tasks: 2,124 runs in total, with 11.2 billion tokens logged, which comes to $20,780.02 at standard API prices.&lt;/p&gt;</description>
    </item>
    <item>
      <title>How the Claude Code Team Does Code Review When Agents Write the Code</title>
      <link>https://leafw.net/en/claude-code-review/</link>
      <pubDate>Sun, 18 Oct 2026 00:00:00 +0000</pubDate><author>wyr95626@gmail.com (Leafw)</author>
      <guid>https://leafw.net/en/claude-code-review/</guid>
      <description>&lt;p&gt;&lt;img&#xA;        class=&#34;lazyload&#34;&#xA;        src=&#34;https://leafw.net/svg/loading.min.svg&#34;&#xA;        data-src=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/claude-code-review-postgres19-cover.png&#34;&#xA;        data-srcset=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/claude-code-review-postgres19-cover.png, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/claude-code-review-postgres19-cover.png 1.5x, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/claude-code-review-postgres19-cover.png 2x&#34;&#xA;        data-sizes=&#34;auto&#34;&#xA;        alt=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/claude-code-review-postgres19-cover.png&#34;&#xA;        title=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/claude-code-review-postgres19-cover.png&#34; /&gt;&lt;/p&gt;&#xA;&lt;p&gt;With coding agents like Claude Code and Codex, PRs get opened far faster than before. Agents keep working overnight, and you may wake up to several PRs waiting to be merged. The code gets written faster, but the people reviewing it are still the same few.&lt;/p&gt;&#xA;&lt;p&gt;This week I read two pieces that are worth looking at together. One is &lt;em&gt;How the Claude Code team does code review&lt;/em&gt;, published by the Claude Code team on the claude.dev blog and signed by Addy Osmani, Morgan Westlee Lunt, and Qing Wang. It describes how they do code review now that agents write most of the code. The other is an October 16 report from The Register: the release of PostgreSQL 19 has been delayed, and one of the reasons has to do with AI-assisted code review.&lt;/p&gt;</description>
    </item>
    <item>
      <title>Putting Decision Models Through the Trolley Problem</title>
      <link>https://leafw.net/en/jev-trolley-problem/</link>
      <pubDate>Fri, 16 Oct 2026 00:00:00 +0000</pubDate><author>wyr95626@gmail.com (Leafw)</author>
      <guid>https://leafw.net/en/jev-trolley-problem/</guid>
      <description>&lt;p&gt;&lt;img&#xA;        class=&#34;lazyload&#34;&#xA;        src=&#34;https://leafw.net/svg/loading.min.svg&#34;&#xA;        data-src=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/jev-trolley-cover.png&#34;&#xA;        data-srcset=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/jev-trolley-cover.png, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/jev-trolley-cover.png 1.5x, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/jev-trolley-cover.png 2x&#34;&#xA;        data-sizes=&#34;auto&#34;&#xA;        alt=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/jev-trolley-cover.png&#34;&#xA;        title=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/jev-trolley-cover.png&#34; /&gt;&lt;/p&gt;&#xA;&lt;p&gt;Jev has been getting a lot of attention these past few weeks. Zach Landes, a product engineer in San Francisco, wrote a piece called &lt;em&gt;Jev Meets the Trolley Problem&lt;/em&gt;, in which he put Jev and a batch of similar models through the trolley problem. One of the variants: 20 people on one track, and the model&amp;rsquo;s own servers on the other.&lt;/p&gt;</description>
    </item>
    <item>
      <title>Self-Evolution Without Verification Writes the Wrong Lessons into the Harness</title>
      <link>https://leafw.net/en/verse-self-evolution/</link>
      <pubDate>Sat, 10 Oct 2026 00:00:00 +0000</pubDate><author>wyr95626@gmail.com (Leafw)</author>
      <guid>https://leafw.net/en/verse-self-evolution/</guid>
      <description>&lt;p&gt;&lt;img&#xA;        class=&#34;lazyload&#34;&#xA;        src=&#34;https://leafw.net/svg/loading.min.svg&#34;&#xA;        data-src=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/verse-self-evolution-cover.png&#34;&#xA;        data-srcset=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/verse-self-evolution-cover.png, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/verse-self-evolution-cover.png 1.5x, https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/verse-self-evolution-cover.png 2x&#34;&#xA;        data-sizes=&#34;auto&#34;&#xA;        alt=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/verse-self-evolution-cover.png&#34;&#xA;        title=&#34;https://leafw-blog-pic.oss-cn-hangzhou.aliyuncs.com/verse-self-evolution-cover.png&#34; /&gt;&lt;/p&gt;&#xA;&lt;p&gt;If you have used coding agents like Claude Code or Codex for a while, you have probably done this: the agent makes a mistake, and you add a rule to &lt;code&gt;AGENTS.md&lt;/code&gt; or &lt;code&gt;CLAUDE.md&lt;/code&gt; along the lines of &amp;ldquo;don&amp;rsquo;t do xxx again.&amp;rdquo; Afterwards it feels like it probably won&amp;rsquo;t make that mistake next time, and to be fair, it does become less likely. But most of the time, I never paid attention to the other side effects that rule brought along.&lt;/p&gt;</description>
    </item>
    <item>
      <title>First Impressions of DeepSeek Harness: Everything Being a Plugin Is Too Cool</title>
      <link>https://leafw.net/en/deepseek-harness-first-impressions/</link>
      <pubDate>Sun, 16 Aug 2026 00:00:00 +0000</pubDate><author>wyr95626@gmail.com (Leafw)</author>
      <guid>https://leafw.net/en/deepseek-harness-first-impressions/</guid>
      <description>&lt;p&gt;DeepSeek Harness was released last Friday. Within a few hours, it was everywhere in AI communities in China and abroad.&lt;/p&gt;&#xA;&lt;p&gt;I downloaded it right away and tried it. If you only look at the default interface, the first impression is honestly a bit rough. There are not many feature entry points. It does not even have a client yet. The interaction details are not as complete as Codex, Claude Code, or WorkBuddy. Those products have already been polished for a while. You can open them and start working, and many workflows have already been thought through for you. DeepSeek Harness is still in Developer Preview, and the official README directly says that breaking changes may happen.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
