<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>StreamFlowMesh Blog</title>
    <link>https://streamflowmesh.io/blog</link>
    <description>Engineering notes on real-time event streaming, event mesh architecture and AI agent infrastructure.</description>
    <language>en</language>
    <lastBuildDate>Thu, 06 Aug 2026 22:14:35 GMT</lastBuildDate>
    <atom:link href="https://streamflowmesh.io/rss.xml" rel="self" type="application/rss+xml"/>
    <atom:link href="https://pubsubhubbub.appspot.com/" rel="hub"/>
    <item>
      <title>Why we rebuilt the Kafka wire protocol from scratch for AI agents</title>
      <link>https://streamflowmesh.io/blog/why-we-rebuilt-kafka-wire-protocol-for-ai-agents</link>
      <guid isPermaLink="true">https://streamflowmesh.io/blog/why-we-rebuilt-kafka-wire-protocol-for-ai-agents</guid>
      <description>Three architectural bets, a benchmark at 4.13M msg/s on commodity hardware, and the one thing we got wrong along the way.</description>
      <category>Engineering</category>
      <pubDate>Thu, 06 Aug 2026 22:14:35 GMT</pubDate>
    </item>
    <item>
      <title>The observe → decide → act loop: an architecture pattern for real-time AI agents</title>
      <link>https://streamflowmesh.io/blog/observe-decide-act-loop-architecture-pattern</link>
      <guid isPermaLink="true">https://streamflowmesh.io/blog/observe-decide-act-loop-architecture-pattern</guid>
      <description>The pattern production AI agent teams keep converging on, why separating the three stages matters, and how to implement it cleanly.</description>
      <category>Architecture</category>
      <pubDate>Thu, 06 Aug 2026 22:14:35 GMT</pubDate>
    </item>
    <item>
      <title>MCP tool calling in production: patterns that survive real traffic</title>
      <link>https://streamflowmesh.io/blog/mcp-tool-calling-production-patterns</link>
      <guid isPermaLink="true">https://streamflowmesh.io/blog/mcp-tool-calling-production-patterns</guid>
      <description>Model Context Protocol works beautifully in a demo. In production, tool calls fail, time out, and cascade. Here are the four patterns we use to keep them boring.</description>
      <category>Engineering</category>
      <pubDate>Thu, 06 Aug 2026 22:14:35 GMT</pubDate>
    </item>
    <item>
      <title>Why stateful AI agents do not need a vector database</title>
      <link>https://streamflowmesh.io/blog/stateful-agents-without-vector-db</link>
      <guid isPermaLink="true">https://streamflowmesh.io/blog/stateful-agents-without-vector-db</guid>
      <description>Vector databases became the default memory layer for AI agents almost by accident. For most agentic workloads, they are the wrong tool. Here is what to use instead.</description>
      <category>Architecture</category>
      <pubDate>Thu, 06 Aug 2026 22:14:35 GMT</pubDate>
    </item>
    <item>
      <title>Event mesh vs message broker: when Kafka is the wrong shape</title>
      <link>https://streamflowmesh.io/blog/event-mesh-vs-message-broker</link>
      <guid isPermaLink="true">https://streamflowmesh.io/blog/event-mesh-vs-message-broker</guid>
      <description>Kafka is a message broker. Event meshes route decisions. The distinction matters more the moment you put AI agents on the wire.</description>
      <category>Architecture</category>
      <pubDate>Thu, 06 Aug 2026 22:14:35 GMT</pubDate>
    </item>
    <item>
      <title>Backpressure for AI agents: the pattern nobody teaches</title>
      <link>https://streamflowmesh.io/blog/backpressure-ai-agents</link>
      <guid isPermaLink="true">https://streamflowmesh.io/blog/backpressure-ai-agents</guid>
      <description>Every streaming system needs backpressure. AI agents make it harder because the cost per message is 1000x higher and unpredictable. Here is the pattern that works.</description>
      <category>Engineering</category>
      <pubDate>Thu, 06 Aug 2026 22:14:35 GMT</pubDate>
    </item>
    <item>
      <title>The Kafka wire protocol is not the hard part — the runtime is</title>
      <link>https://streamflowmesh.io/blog/kafka-wire-protocol-runtime</link>
      <guid isPermaLink="true">https://streamflowmesh.io/blog/kafka-wire-protocol-runtime</guid>
      <description>Reimplementing the Kafka wire protocol is a solved problem. The hard part is what happens after the bytes arrive. Here is what a modern streaming runtime has to get right.</description>
      <category>Engineering</category>
      <pubDate>Thu, 06 Aug 2026 22:14:35 GMT</pubDate>
    </item>
    <item>
      <title>Why Most Enterprise AI Agent Platforms Fail at Scale</title>
      <link>https://streamflowmesh.io/blog/why-most-enterprise-ai-agent-platforms-fail-at-scale</link>
      <guid isPermaLink="true">https://streamflowmesh.io/blog/why-most-enterprise-ai-agent-platforms-fail-at-scale</guid>
      <description>Most enterprise AI agent platforms break when they hit production scale. Here&apos;s why and what a real-time decision engine does differently.</description>
      <category>Engineering</category>
      <pubDate>Sat, 11 Apr 2026 21:54:42 GMT</pubDate>
    </item>
    <item>
      <title>How AI Agents are Transforming Real-Time Data Streaming</title>
      <link>https://streamflowmesh.io/blog/how-ai-agents-are-transforming-real-time-data-streaming</link>
      <guid isPermaLink="true">https://streamflowmesh.io/blog/how-ai-agents-are-transforming-real-time-data-streaming</guid>
      <description>AI agents are revolutionizing real-time data streaming, moving beyond simple collection to proactive understanding and reaction. Discover how StreamFlow enables this transformation.</description>
      <category>Engineering</category>
      <pubDate>Sat, 11 Apr 2026 21:26:36 GMT</pubDate>
    </item>
  </channel>
</rss>
