<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>structured-data on S Anand</title>
    <link>https://www.s-anand.net/blog/tag/structured-data/</link>
    <description>Recent content in structured-data on S Anand</description>
    <generator>Hugo -- 0.164.0</generator>
    <language>en-us</language>
    <lastBuildDate>Sat, 11 Jul 2026 21:24:39 +0530</lastBuildDate>
    <atom:link href="https://www.s-anand.net/blog/tag/structured-data/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>When Data is for Agents - Workshop Summary</title>
      <link>https://www.s-anand.net/blog/when-data-is-for-agents-workshop-summary/</link>
      <pubDate>Sat, 11 Jul 2026 21:24:39 +0530</pubDate>
      <guid>https://www.s-anand.net/blog/when-data-is-for-agents-workshop-summary/</guid>
      <description>&lt;p&gt;Here&amp;rsquo;s roughly what I said in my &lt;a href=&#34;https://hasgeek.com/fifthelephant/when-data-is-for-agents-workshop/&#34;&gt;When Data is for Agents&lt;/a&gt; workshop for &lt;a href=&#34;https://hasgeek.com/fifthelephant/&#34;&gt;Fifth Elephant&lt;/a&gt; on 7 Jul 2026.&lt;/p&gt;
&lt;p&gt;Or you can read the &lt;a href=&#34;https://sanand0.github.io/talks/2026-07-07-when-data-is-for-agents-fifth-elephant/&#34;&gt;detailed AI-generated version&lt;/a&gt; if you prefer - it has all the prompts, links, results, etc.&lt;/p&gt;
&lt;hr&gt;
&lt;p&gt;I think agents prefer data in a different form than humans. But I don&amp;rsquo;t know.&lt;/p&gt;
&lt;p&gt;So, everyone, open ChatGPT (or Claude or whatever), &lt;a href=&#34;https://github.com/sanand0/talks/blob/226a6208889bf3d2bb1fb24d923d9298ec3fc81d/2026-07-07-when-data-is-for-agents-fifth-elephant/research-prompt.md&#34;&gt;research&lt;/a&gt; and ask it!&lt;/p&gt;
&lt;p&gt;Now, let&amp;rsquo;s &lt;a href=&#34;https://github.com/sanand0/talks/blob/226a6208889bf3d2bb1fb24d923d9298ec3fc81d/2026-07-07-when-data-is-for-agents-fifth-elephant/collation-prompt.md&#34;&gt;collate them&lt;/a&gt; and &lt;a href=&#34;https://chatgpt.com/share/6a525953-3da0-83e8-94a5-a8d99b291fc7&#34;&gt;see the result&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;Aha! Looks like:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Progressive, just-in-time access beats feeding full context at once&lt;/li&gt;
&lt;li&gt;For changing, text-heavy corpora, grep-style navigation can beat embeddings&lt;/li&gt;
&lt;li&gt;CSV beats JSON on accuracy per token&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;Hmm&amp;hellip; let&amp;rsquo;s test them one by one.&lt;/p&gt;
&lt;p&gt;Everyone, ask Codex (or Claude Code) to &lt;a href=&#34;https://github.com/sanand0/talks/blob/226a6208889bf3d2bb1fb24d923d9298ec3fc81d/2026-07-07-when-data-is-for-agents-fifth-elephant/benchmark-prompt.md&#34;&gt;run a benchmark&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;What? All thirteen of your results said, &amp;ldquo;It makes no difference?&amp;rdquo; Huh&amp;hellip;&lt;/p&gt;
&lt;p&gt;Claude, &lt;a href=&#34;https://claude.ai/share/19d72406-b262-482e-895e-c7da0fde3382&#34;&gt;did my benchmarking prompt do a good job&lt;/a&gt;?&lt;/p&gt;
&lt;p&gt;What? I messed up my benchmarking prompt? It gave all the data at once instead of progressively? OK, give me the revised prompt.&lt;/p&gt;
&lt;p&gt;OK, everyone, try this prompt. What does it say?&lt;/p&gt;
&lt;p&gt;Aha! Yes, progressive disclosure costs 2.5x - 25x less for the same accuracy. Cool!&lt;/p&gt;
&lt;p&gt;Let&amp;rsquo;s try another experiment. What? I totally messed up that prompt as well? Er&amp;hellip; we&amp;rsquo;re short of time?&lt;/p&gt;
&lt;p&gt;Fine, wrap up, then. Claude, convert what we learnt into a &lt;a href=&#34;https://github.com/sanand0/talks/blob/226a6208889bf3d2bb1fb24d923d9298ec3fc81d/2026-07-07-when-data-is-for-agents-fifth-elephant/corpus-for-agents-skill.md&#34;&gt;skill&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;OK, folks, my takeaways:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;I don&amp;rsquo;t know what formats agents prefer, but we can ask AI to research.&lt;/li&gt;
&lt;li&gt;I don&amp;rsquo;t know if the research is right, but we can ask AI to create a benchmark.&lt;/li&gt;
&lt;li&gt;I don&amp;rsquo;t know if the benchmark is right, but we can ask AI to audit it.&lt;/li&gt;
&lt;li&gt;I don&amp;rsquo;t know if I&amp;rsquo;ll remember what&amp;rsquo;s valid, so we can ask AI to create and reuse skills.&lt;/li&gt;
&lt;/ol&gt;
&lt;hr&gt;
&lt;p&gt;&lt;img loading=&#34;lazy&#34; src=&#34;https://sanand0.github.io/talks/2026-07-07-when-data-is-for-agents-fifth-elephant/summary.avif&#34;&gt;&lt;/p&gt;
</description>
    </item>
    <item>
      <title>Openings at Google</title>
      <link>https://www.s-anand.net/blog/openings-at-google/</link>
      <pubDate>Wed, 29 Mar 2006 12:00:00 +0000</pubDate>
      <guid>https://www.s-anand.net/blog/openings-at-google/</guid>
      <description>&lt;p&gt;&lt;a href=&#34;http://www.google.com/base/search?q=&amp;amp;oi=gb_refinement&amp;amp;ct=t&amp;amp;cd=1&amp;amp;a_n0=jobs&amp;amp;a_y0=9&amp;amp;a_o0=0&amp;amp;a_n152=Location&amp;amp;a_y152=6&amp;amp;a_v152=&amp;amp;a_o152=0&amp;amp;a_n759=Employer&amp;amp;a_y759=1&amp;amp;a_v759=google%20&amp;amp;a_o759=0&amp;amp;a_n215=Job%20function&amp;amp;a_y215=1&#34;&gt;Job openings at Google&lt;/a&gt; on Google Base.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id=&#34;comments&#34;&gt;Comments&lt;/h2&gt;
&lt;!-- wp-comments-start --&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Gautam&lt;/strong&gt; &lt;em&gt;30 Mar 2006 6:39 am&lt;/em&gt;:
Yeah, naukri, Monster and Recruitment Management Systems need to spruce up their offerings&lt;/li&gt;
&lt;/ul&gt;
&lt;!-- wp-comments-end --&gt;
</description>
    </item>
    <item>
      <title>MIT OpenCourseWare</title>
      <link>https://www.s-anand.net/blog/mit-opencourseware/</link>
      <pubDate>Wed, 16 Nov 2005 12:00:00 +0000</pubDate>
      <guid>https://www.s-anand.net/blog/mit-opencourseware/</guid>
      <description>&lt;p&gt;&lt;a href=&#34;http://base.google.com/base/search?authorid=314965&#34;&gt;MIT OpenCourseWare&lt;/a&gt; on Google Base.&lt;/p&gt;
</description>
    </item>
    <item>
      <title>Lists to Maps</title>
      <link>https://www.s-anand.net/blog/lists-to-maps/</link>
      <pubDate>Sun, 04 Sep 2005 12:00:00 +0000</pubDate>
      <guid>https://www.s-anand.net/blog/lists-to-maps/</guid>
      <description>&lt;p&gt;How to &lt;a href=&#34;http://www.hackdiary.com/archives/000070.html&#34;&gt;turn text lists into structured maps using the Web&lt;/a&gt;.&lt;/p&gt;
</description>
    </item>
    <item>
      <title>Semantic web</title>
      <link>https://www.s-anand.net/blog/semantic-web/</link>
      <pubDate>Sat, 27 Jul 2002 12:00:00 +0000</pubDate>
      <guid>https://www.s-anand.net/blog/semantic-web/</guid>
      <description>&lt;p&gt;A story dreaming for the &lt;a href=&#34;http://ftrain.com/google_takes_all.html&#34;&gt;semantic web&lt;/a&gt;.&lt;/p&gt;
</description>
    </item>
  </channel>
</rss>
