<?xml version="1.0" encoding="UTF-8" ?>
<rss version="2.0">
    <channel>
      <title>Floris Vossebeld</title>
      <link>https://vossebeld.dev</link>
      <description>Last 10 notes on Floris Vossebeld</description>
      <generator>Quartz -- quartz.jzhao.xyz</generator>
      <item>
    <title>Agent workspaces</title>
    <link>https://vossebeld.dev/concepts/agent-workspaces</link>
    <guid>https://vossebeld.dev/concepts/agent-workspaces</guid>
    <description><![CDATA[ Capability lives in the coupling between a model and its workspace. Give it a durable environment and behaviour changes; the two co-evolve. ]]></description>
    <pubDate>Tue, 06 Oct 2026 10:17:36 GMT</pubDate>
  </item><item>
    <title>Context window</title>
    <link>https://vossebeld.dev/concepts/context-window</link>
    <guid>https://vossebeld.dev/concepts/context-window</guid>
    <description><![CDATA[ The active space an agent reasons in. The hard part is selection, not size; a connected data source is not live context. ]]></description>
    <pubDate>Tue, 06 Oct 2026 10:17:36 GMT</pubDate>
  </item><item>
    <title>Federated memory for enterprise agents</title>
    <link>https://vossebeld.dev/concepts/federated-memory-for-enterprise-agents</link>
    <guid>https://vossebeld.dev/concepts/federated-memory-for-enterprise-agents</guid>
    <description><![CDATA[ Letting agents learn from experience without leaking across users, teams, projects, or customers: a taxonomy, a threat model, and the metadata every memory needs. ]]></description>
    <pubDate>Tue, 06 Oct 2026 10:17:36 GMT</pubDate>
  </item><item>
    <title>The memory promotion pipeline</title>
    <link>https://vossebeld.dev/concepts/memory-promotion-pipeline</link>
    <guid>https://vossebeld.dev/concepts/memory-promotion-pipeline</guid>
    <description><![CDATA[ How raw agent experience should move through abstraction and review before it becomes shared knowledge, with a worked example and its failure modes. ]]></description>
    <pubDate>Tue, 06 Oct 2026 10:17:36 GMT</pubDate>
  </item><item>
    <title>Orchestrating scoped agents</title>
    <link>https://vossebeld.dev/concepts/orchestrating-scoped-agents</link>
    <guid>https://vossebeld.dev/concepts/orchestrating-scoped-agents</guid>
    <description><![CDATA[ Scoping an agent to one system makes it safe but leaves it unable to finish cross-system work. The fix is broad intent, narrow execution, and the hard part is owning the half-finished state when a step fails. ]]></description>
    <pubDate>Tue, 06 Oct 2026 10:17:36 GMT</pubDate>
  </item><item>
    <title>Scoped system-specialist agents</title>
    <link>https://vossebeld.dev/concepts/scoped-system-specialist-agents</link>
    <guid>https://vossebeld.dev/concepts/scoped-system-specialist-agents</guid>
    <description><![CDATA[ Why enterprise agents should be scoped to one system rather than one broad autonomous operator, and why scope alone isn't enough. ]]></description>
    <pubDate>Tue, 06 Oct 2026 10:17:36 GMT</pubDate>
  </item><item>
    <title>The agent as semantic UI</title>
    <link>https://vossebeld.dev/concepts/the-agent-as-semantic-ui</link>
    <guid>https://vossebeld.dev/concepts/the-agent-as-semantic-ui</guid>
    <description><![CDATA[ A semantic UI is a control surface that does the thing. A chatbot is a help surface that tells you where to do it. The distinction is load-bearing for enterprise agents. ]]></description>
    <pubDate>Tue, 06 Oct 2026 10:17:36 GMT</pubDate>
  </item><item>
    <title>How this site works</title>
    <link>https://vossebeld.dev/how-this-works</link>
    <guid>https://vossebeld.dev/how-this-works</guid>
    <description><![CDATA[ The living-wiki / agent-memory idea behind this digital garden. ]]></description>
    <pubDate>Tue, 06 Oct 2026 10:17:36 GMT</pubDate>
  </item><item>
    <title>From chatbots to system operators</title>
    <link>https://vossebeld.dev/thoughts/from-chatbots-to-system-operators</link>
    <guid>https://vossebeld.dev/thoughts/from-chatbots-to-system-operators</guid>
    <description><![CDATA[ A working theory about how LLM agents are evolving from chat partners into governed operators of real systems, and the architecture and memory problems that come with it. ]]></description>
    <pubDate>Tue, 06 Oct 2026 10:17:36 GMT</pubDate>
  </item><item>
    <title>Agents reach for human tools, not machine protocols</title>
    <link>https://vossebeld.dev/thoughts/human-tools-not-machine-protocols</link>
    <guid>https://vossebeld.dev/thoughts/human-tools-not-machine-protocols</guid>
    <description><![CDATA[ When we give models the ability to act, the interfaces that stick are the ones humans built for humans, not the ones machines built to talk to each other. ]]></description>
    <pubDate>Tue, 06 Oct 2026 10:17:36 GMT</pubDate>
  </item>
    </channel>
  </rss>