<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Field Notes — Taanis Karigar</title>
    <link>https://taanis.nl/blog</link>
    <atom:link href="https://taanis.nl/feed.xml" rel="self" type="application/rss+xml"/>
    <description>Long-form notes on data engineering, AI experiments, and building systems that actually ship.</description>
    <language>en</language>
    <lastBuildDate>Mon, 02 Nov 2026 10:03:15 +0100</lastBuildDate>
    <item>
      <title>Why my coding agents moved from panes to threads</title>
      <link>https://taanis.nl/bb-agent-workbench-field-note-09</link>
      <guid isPermaLink="true">https://taanis.nl/bb-agent-workbench-field-note-09</guid>
      <pubDate>Mon, 02 Nov 2026 10:03:15 +0100</pubDate>
      <description>Four agents in herdr at the desk. A thread, a diff and a place to come back to in bb. What changed, what it costs, and where the terminal is still better.</description>
    </item>
    <item>
      <title>The AI coding setup that caught its own mistakes</title>
      <link>https://taanis.nl/ai-coding-orchestration-field-note-08</link>
      <guid isPermaLink="true">https://taanis.nl/ai-coding-orchestration-field-note-08</guid>
      <pubDate>Wed, 12 Aug 2026 02:54:16 -0400</pubDate>
      <description>One planner, one builder and one critic in visible terminals, with written handoffs and a person owning the release gate. This is what building Docker support for Polster taught me about reliable AI coding.</description>
    </item>
    <item>
      <title>DuckLake is the new Delta Lake</title>
      <link>https://taanis.nl/ducklake-is-the-new-delta-lake-field-note-07</link>
      <guid isPermaLink="true">https://taanis.nl/ducklake-is-the-new-delta-lake-field-note-07</guid>
      <pubDate>Mon, 13 Jul 2026 18:47:15 +0200</pubDate>
      <description>The same open-lakehouse job Delta made mainstream, with a different catalog and a real gap in access control. A working note on what the claim means and where it falls apart.</description>
    </item>
    <item>
      <title>Replace Your SQL Pipelines with Polars</title>
      <link>https://taanis.nl/polars-replace-sql-pipelines-field-note-06</link>
      <guid isPermaLink="true">https://taanis.nl/polars-replace-sql-pipelines-field-note-06</guid>
      <pubDate>Sun, 05 Jul 2026 20:17:14 +0200</pubDate>
      <description>Your transformation stack is probably four tools and three languages. A Polars, Dagster and pytest stack is one language for any data source, with data contracts you can enforce in CI.</description>
    </item>
    <item>
      <title>SQL or Python for Data Transformations? Start With the Engine</title>
      <link>https://taanis.nl/sql-python-transformations-field-note-05</link>
      <guid isPermaLink="true">https://taanis.nl/sql-python-transformations-field-note-05</guid>
      <pubDate>Sun, 05 Jul 2026 20:17:14 +0200</pubDate>
      <description>SQL won the transformation layer and Python won the system around it. The useful question is where the data lives, what the transform does and who has to maintain it.</description>
    </item>
    <item>
      <title>Your Databricks Lakehouse Is a Postgres Table and a Parquet Folder</title>
      <link>https://taanis.nl/ducklake-open-lakehouse-field-note-04</link>
      <guid isPermaLink="true">https://taanis.nl/ducklake-open-lakehouse-field-note-04</guid>
      <pubDate>Fri, 26 Jun 2026 18:01:33 +0200</pubDate>
      <description>Most of what a lakehouse does comes down to two things most companies already know how to run, and the price difference is pretty big.</description>
    </item>
    <item>
      <title>A Hetzner Box vs. a Databricks cluster</title>
      <link>https://taanis.nl/hetzner-vs-databricks</link>
      <guid isPermaLink="true">https://taanis.nl/hetzner-vs-databricks</guid>
      <pubDate>Sat, 13 Jun 2026 20:03:03 +0200</pubDate>
      <description>Most companies are paying for a fleet of trucks to deliver a single envelope. Here&#x27;s the cheaper way, and when it stops working.</description>
    </item>
    <item>
      <title>Three CEOs on what work becomes</title>
      <link>https://taanis.nl/three-ceos-future-of-work</link>
      <guid isPermaLink="true">https://taanis.nl/three-ceos-future-of-work</guid>
      <pubDate>Sat, 13 Jun 2026 08:52:04 +0200</pubDate>
      <description>Same direction, different readings. A research note on the white-collar question.</description>
    </item>
    <item>
      <title>Why Multi-Agent Systems Work</title>
      <link>https://taanis.nl/why-multi-agent-systems</link>
      <guid isPermaLink="true">https://taanis.nl/why-multi-agent-systems</guid>
      <pubDate>Tue, 02 Jun 2026 14:27:36 +0200</pubDate>
      <description>A single large model, asked to do four jobs at once, will blend them. The fix is structural.</description>
    </item>
  </channel>
</rss>
