<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title><![CDATA[Terry Murray — Reading Log]]></title>
    <description><![CDATA[AI content Terry Murray is reading and bookmarking.]]></description>
    <link>https://terrymurray.com</link>
    <atom:link href="https://terrymurray.com/feed/log" rel="self" type="application/rss+xml"/>
    <language>en-us</language>
    <lastBuildDate>Thu, 16 Apr 2026 11:07:44 GMT</lastBuildDate>
    <item>
      <title><![CDATA[Claude Mythos Preview \ red.anthropic.com]]></title>
      <link>https://red.anthropic.com/2026/mythos-preview/</link>
      <guid isPermaLink="false">https://terrymurray.com/post/c8c2ccc8-d723-45ce-a01f-c715a3d21f61</guid>
      <pubDate>Wed, 08 Apr 2026 12:17:49 GMT</pubDate>
      <description><![CDATA[[red.anthropic.com] Given how good this system is at security, I’m excited to see how it does in large code bases.]]></description>
    </item>
    <item>
      <title><![CDATA[Amazon S3 Files]]></title>
      <link>https://aws.amazon.com/s3/features/files/</link>
      <guid isPermaLink="false">https://terrymurray.com/post/3d7855ac-3ed7-4973-9e5d-2eb838ba38d2</guid>
      <pubDate>Wed, 08 Apr 2026 02:07:21 GMT</pubDate>
      <description><![CDATA[[aws.amazon.com] This is the way to go if you have a decent sized user base (say a 500 person sized dev team) using agents against the same file based data store.]]></description>
    </item>
    <item>
      <title><![CDATA[Google Maps for Codebases: Paste a GitHub URL, Ask Anything - DEV Community]]></title>
      <link>https://dev.to/copilotkit/google-maps-for-codebases-paste-a-github-url-ask-anything-3hk8</link>
      <guid isPermaLink="false">https://terrymurray.com/post/ac964538-de8d-4374-a4d3-a605ea360eb1</guid>
      <pubDate>Tue, 07 Apr 2026 21:18:45 GMT</pubDate>
      <description><![CDATA[[dev.to] Scanning only dependency files and imports at the tops of code files is a great way to cheaply (inference) build a code map.]]></description>
    </item>
    <item>
      <title><![CDATA[Claude Code leak]]></title>
      <link>https://www.theverge.com/ai-artificial-intelligence/904776/anthropic-claude-source-code-leak</link>
      <guid isPermaLink="false">https://terrymurray.com/post/682d2d6c-2d24-4566-9871-88b68274f2ac</guid>
      <pubDate>Wed, 01 Apr 2026 11:34:08 GMT</pubDate>
      <description><![CDATA[[www.theverge.com] This is all fascinating.]]></description>
    </item>
    <item>
      <title><![CDATA[GitHub - everything-claude-code: The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development]]></title>
      <link>https://github.com/affaan-m/everything-claude-code</link>
      <guid isPermaLink="false">https://terrymurray.com/post/bccbaa44-823b-4253-84d7-4f58ff51c7ec</guid>
      <pubDate>Wed, 01 Apr 2026 03:02:36 GMT</pubDate>
      <description><![CDATA[[github.com] This is worth pouring over.]]></description>
    </item>
    <item>
      <title><![CDATA[Agent Evaluation Readiness Checklist]]></title>
      <link>https://blog.langchain.com/agent-evaluation-readiness-checklist/</link>
      <guid isPermaLink="false">https://terrymurray.com/post/9bb2d67c-6743-473d-8970-0bd49507ab76</guid>
      <pubDate>Sun, 29 Mar 2026 20:07:06 GMT</pubDate>
      <description><![CDATA[[blog.langchain.com] Practical guidelines for evaluating autonomous agents.]]></description>
    </item>
    <item>
      <title><![CDATA[Anthropic Just Dropped the Feature Nobody Knew They Needed - YouTube]]></title>
      <link>https://www.youtube.com/watch?v=OnQ4BGN8B-s</link>
      <guid isPermaLink="false">https://terrymurray.com/post/8f7f5ff3-6c68-4b20-b84e-480fb56bf16b</guid>
      <pubDate>Wed, 25 Mar 2026 01:41:19 GMT</pubDate>
      <description><![CDATA[[www.youtube.com] Anthropic has added a glymphatic system to Claude.]]></description>
    </item>
    <item>
      <title><![CDATA[karpathy/autoresearch: AI agents running research on single-GPU nanochat training automatically]]></title>
      <link>https://github.com/karpathy/autoresearch</link>
      <guid isPermaLink="false">https://terrymurray.com/post/f91f239e-44cf-4cde-960f-4173fed017d9</guid>
      <pubDate>Tue, 17 Mar 2026 23:08:25 GMT</pubDate>
      <description><![CDATA[[github.com] A friend recommend this (hi David!). What's important here is that if you can setup an eval that an LLM can use to evaluate it's work, and then give it some parameter space it's permitted to modify, it can experiment with those parameters, eval, and iteratively improve the system under test without further human involvement.]]></description>
    </item>
    <item>
      <title><![CDATA[Developers Watching their Non-Tech Founder Vibe Coding a New Feature #de...]]></title>
      <link>https://youtube.com/shorts/toVYr9MC_1s?si=WGc0o5IlUsT31u7I</link>
      <guid isPermaLink="false">https://terrymurray.com/post/af5a4ec8-a34e-4b05-bc29-eb61d899e061</guid>
      <pubDate>Fri, 13 Mar 2026 23:13:15 GMT</pubDate>
      <description><![CDATA[[youtube.com] OMG I'm dying. His face.]]></description>
    </item>
    <item>
      <title><![CDATA[In wake of outage, Amazon calls upon senior engineers to address issues created by 'Gen-AI assisted changes,' report claims — recent 'high blast radius' incidents stir up changes for code approval | Tom's Hardware]]></title>
      <link>https://www.tomshardware.com/tech-industry/artificial-intelligence/amazon-calls-engineers-to-address-issues-caused-by-use-of-ai-tools-report-claims-company-says-recent-incidents-had-high-blast-radius-and-were-allegedly-related-to-gen-ai-assisted-changes</link>
      <guid isPermaLink="false">https://terrymurray.com/post/65d00ce9-c5cc-40cf-ad76-b091230f2c83</guid>
      <pubDate>Wed, 11 Mar 2026 11:01:29 GMT</pubDate>
      <description><![CDATA[[www.tomshardware.com] Quality control must be the next major focus in both LLM assisted coding and agentic coding.]]></description>
    </item>
    <item>
      <title><![CDATA[Code Review for Claude Code | Claude]]></title>
      <link>https://claude.com/blog/code-review</link>
      <guid isPermaLink="false">https://terrymurray.com/post/33c53b3c-8ea3-4063-ba45-ef3f6a00a462</guid>
      <pubDate>Tue, 10 Mar 2026 00:15:39 GMT</pubDate>
      <description><![CDATA[[claude.com] Now THIS is interesting.]]></description>
    </item>
    <item>
      <title><![CDATA[The case for running AI agents on Markdown files instead of MCP servers - The New Stack]]></title>
      <link>https://thenewstack.io/skills-vs-mcp-agent-architecture/</link>
      <guid isPermaLink="false">https://terrymurray.com/post/65958a12-43a2-49ff-943e-04c432e3996c</guid>
      <pubDate>Sat, 07 Mar 2026 17:42:00 GMT</pubDate>
      <description><![CDATA[[thenewstack.io] When to use an MCP server, vs. when to use a markdown skill.]]></description>
    </item>
    <item>
      <title><![CDATA[Tell HN: I'm 60 years old. Claude Code has re-ignited a passion | Hacker News]]></title>
      <link>https://news.ycombinator.com/item?id=47282777</link>
      <guid isPermaLink="false">https://terrymurray.com/post/ca0a76b7-6d48-4195-ade3-1a93625cfb30</guid>
      <pubDate>Sat, 07 Mar 2026 17:19:09 GMT</pubDate>
      <description><![CDATA[[news.ycombinator.com] These are my people. I'm equally fed up with having to relearn, over and over, new frameworks that accomplish the same outcomes as the tools I used 30 years ago. LLMs have made it fun again.]]></description>
    </item>
    <item>
      <title><![CDATA[generative-ai/gemini/agents/always-on-memory-agent at main · GoogleCloudPlatform/generative-ai · GitHub]]></title>
      <link>https://share.google/pRib6AwCKVdFbxQ1J</link>
      <guid isPermaLink="false">https://terrymurray.com/post/f87ba5a7-c8f6-46f9-9f52-fcbe5c3d0364</guid>
      <pubDate>Sat, 07 Mar 2026 02:45:03 GMT</pubDate>
      <description><![CDATA[[share.google] An interesting way to give agents memory outside of vector databases. The 'always on' 30 min cycle idea is interesting.]]></description>
    </item>
    <item>
      <title><![CDATA[Introducing OpenClaw on Amazon Lightsail to run your autonomous private AI agents | AWS News Blog]]></title>
      <link>https://share.google/kBAoXbqknn9PPfwd9</link>
      <guid isPermaLink="false">https://terrymurray.com/post/7ad6f8eb-97ba-4498-8aac-3bd1cd3c83d7</guid>
      <pubDate>Thu, 05 Mar 2026 21:37:48 GMT</pubDate>
      <description><![CDATA[[share.google] You have got to be kidding me.]]></description>
    </item>
    <item>
      <title><![CDATA[How “Deep Industry Research Agents” Can Change Your Organization]]></title>
      <link>https://hbr.org/2026/03/how-deep-industry-research-agents-can-change-your-organization</link>
      <guid isPermaLink="false">https://terrymurray.com/post/9e013e11-4fb3-475d-9f29-4e7724eafed3</guid>
      <pubDate>Wed, 04 Mar 2026 03:15:56 GMT</pubDate>
      <description><![CDATA[[hbr.org] Give agents access to all the data in real workflows, with a memory to learn over time. Let them make recommendations at first, aiming to reach super human ability when they can make real time decisions.]]></description>
    </item>
    <item>
      <title><![CDATA[Anthropic Courses]]></title>
      <link>https://share.google/0OZtf3ppc6lRDqh6x</link>
      <guid isPermaLink="false">https://terrymurray.com/post/6f3c5f46-399f-423c-819f-21dc39f38901</guid>
      <pubDate>Tue, 03 Mar 2026 03:08:13 GMT</pubDate>
      <description><![CDATA[[share.google] Anthropic now has official courses for learning how to use LLMs.]]></description>
    </item>
    <item>
      <title><![CDATA[Cognitive Debt: When Velocity Exceeds Comprehension | rockoder]]></title>
      <link>https://www.rockoder.com/beyondthecode/cognitive-debt-when-velocity-exceeds-comprehension/</link>
      <guid isPermaLink="false">https://terrymurray.com/post/458fde7a-c1f4-4d0b-bd3b-c0f322224e70</guid>
      <pubDate>Mon, 02 Mar 2026 14:54:42 GMT</pubDate>
      <description><![CDATA[[www.rockoder.com] When AI-driven velocity outpaces human understanding, teams accumulate cognitive debt: code that works but isn’t truly comprehended, making systems fragile, harder to maintain, and riskier over time. If organizations optimize purely for output instead of comprehension, they trade short-term speed for long-term instability, slower learning, and declining engineering judgment.]]></description>
    </item>
    <item>
      <title><![CDATA[Boris Cherny on X: "In the next version of Claude Code.. We're introducing two new Skills: /simplify and /batch. I have been using both daily, and am excited to share them with everyone. Combined, these kills automate much of the work it used to take to (1) shepherd a pull request to production https://t.co/mztx6KbDh9" / X]]></title>
      <link>https://t.co/mztx6KbDh9&quot;</link>
      <guid isPermaLink="false">https://terrymurray.com/post/891d547a-de9c-4d23-83ea-860c536c817d</guid>
      <pubDate>Sat, 28 Feb 2026 02:37:35 GMT</pubDate>
      <description><![CDATA[[t.co] /simplify and /batch. I continue to be impressed by how in tune Anthropic is to the needs of devs at the bleeding edge, I assume because they are eating their own cooking.]]></description>
    </item>
    <item>
      <title><![CDATA[I vibe coded my dream macOS presentation app]]></title>
      <link>https://share.google/pCITD7Z4zlPJkUS2p</link>
      <guid isPermaLink="false">https://terrymurray.com/post/ad14c1bf-9d96-4198-b761-3ed278302e72</guid>
      <pubDate>Thu, 26 Feb 2026 12:08:04 GMT</pubDate>
      <description><![CDATA[[share.google] Simon's story is the same story as where TerryMurray.com came from.]]></description>
    </item>
    <item>
      <title><![CDATA[The Future of Software Engineering with AI: Six Predictions]]></title>
      <link>https://newsletter.pragmaticengineer.com/p/the-future-of-software-engineering-with-ai</link>
      <guid isPermaLink="false">https://terrymurray.com/post/6e3df2fa-b234-4a56-b69b-294911539e2e</guid>
      <pubDate>Wed, 25 Feb 2026 13:45:21 GMT</pubDate>
      <description><![CDATA[[newsletter.pragmaticengineer.com] A collection of eclectic margins from industry leaders. Great read of you want to understand how to transform business with AI.]]></description>
    </item>
    <item>
      <title><![CDATA[Harness engineering: leveraging Codex in an agent-first world | OpenAI]]></title>
      <link>https://openai.com/index/harness-engineering/</link>
      <guid isPermaLink="false">https://terrymurray.com/post/34f4792b-b4e3-4b30-98fd-9a1230a18337</guid>
      <pubDate>Sun, 22 Feb 2026 13:04:29 GMT</pubDate>
      <description><![CDATA[[openai.com] I believe this is the future of development.]]></description>
    </item>
    <item>
      <title><![CDATA[How LLMs Actually Generate Text (Every Dev Should Know This) - YouTube]]></title>
      <link>https://www.youtube.com/watch?v=NKnZYvZA7w4</link>
      <guid isPermaLink="false">https://terrymurray.com/post/c30b4472-beb1-42fe-818b-dfb4e6e2d69d</guid>
      <pubDate>Fri, 20 Feb 2026 15:05:43 GMT</pubDate>
      <description><![CDATA[[www.youtube.com] I recommend this look behind the curtain of LLMs. I knew some of this, but learned a ton. Great video.]]></description>
    </item>
    <item>
      <title><![CDATA[Gemini 3.1 Pro | Hacker News]]></title>
      <link>https://news.ycombinator.com/item?id=47075318</link>
      <guid isPermaLink="false">https://terrymurray.com/post/de5a2e1a-e47d-4e33-82a3-0700074ca161</guid>
      <pubDate>Thu, 19 Feb 2026 17:45:06 GMT</pubDate>
      <description><![CDATA[[news.ycombinator.com] Gemini 3.1 pro dropped. It reaches 80.6% on the SWE Bench, where Opus 4.6 gets 80.8%.]]></description>
    </item>
    <item>
      <title><![CDATA[Dario Amodei — “We are near the end of the exponential”]]></title>
      <link>https://youtube.com/watch?v=n1E9IZfvGMA&amp;si=Pa-cuz5_01Rxv34p</link>
      <guid isPermaLink="false">https://terrymurray.com/post/51b50b9c-fa9b-488b-9ccb-cc22305bc9bf</guid>
      <pubDate>Thu, 19 Feb 2026 17:24:20 GMT</pubDate>
      <description><![CDATA[[youtube.com] Dario (CEO of Anthropic) makes a prediction that there's a lot of juice left in RL (reinforcement learning), so we can expect further model improvements (vs being plateaued).]]></description>
    </item>
  </channel>
</rss>