<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Tool-Routing on Truth-First Beacon — Paul Desai</title><link>https://beacon.activemirror.ai/tags/tool-routing/</link><description>Recent content in Tool-Routing on Truth-First Beacon — Paul Desai</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sat, 28 Feb 2026 14:55:00 +0530</lastBuildDate><atom:link href="https://beacon.activemirror.ai/tags/tool-routing/feed.xml" rel="self" type="application/rss+xml"/><item><title>The Fallback Chain: Provider-Agnostic Tool Routing for AI Agents</title><link>https://beacon.activemirror.ai/reflections/the-fallback-chain/</link><pubDate>Sat, 28 Feb 2026 14:55:00 +0530</pubDate><guid>https://beacon.activemirror.ai/reflections/the-fallback-chain/</guid><description>&lt;p&gt;&lt;a href="https://github.com/anomalyco/opencode/issues/10704"&gt;OpenCode issue #10704&lt;/a&gt; landed this week: &lt;em&gt;&amp;ldquo;Use provider-hosted web search when available.&amp;rdquo;&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;The request is specific and correct. Web search in most AI agents today is hardcoded: Exa integration, custom flag, API key. Even when the provider — Claude, OpenAI, Gemini — already ships a native search tool. The developer pays for Exa, configures the key, and gets worse results than what the provider offers natively, because the provider&amp;rsquo;s search is model-native, not bolted on.&lt;/p&gt;</description></item></channel></rss>