<?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>agents on Field Notes</title><link>https://blog.penmen.net/tags/agents/</link><description>Recent content in agents on Field Notes</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Tue, 07 Jul 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://blog.penmen.net/tags/agents/index.xml" rel="self" type="application/rss+xml"/><item><title>Building a Text-to-SQL Agent That Actually Works on Enterprise Data</title><link>https://blog.penmen.net/2026/07/building-a-text-to-sql-agent-that-actually-works-on-enterprise-data/</link><pubDate>Tue, 07 Jul 2026 00:00:00 +0000</pubDate><guid>https://blog.penmen.net/2026/07/building-a-text-to-sql-agent-that-actually-works-on-enterprise-data/</guid><description>LLMs score 85%+ on public SQL benchmarks but collapse to 39–61% on real enterprise schemas. I built a multi-stage data agent for a life insurance data warehouse to close that gap — here&amp;#39;s what 18 months of iteration taught me about why the approach matters more than the model.</description></item></channel></rss>