Updated
Updated · InfoWorld · Aug 26
New XMLUI How-To Lifts MCP Search Score to 4.2 After Agent Missed Radio-Group Task
Updated
Updated · InfoWorld · Aug 26

New XMLUI How-To Lifts MCP Search Score to 4.2 After Agent Missed Radio-Group Task

1 articles · Updated · InfoWorld · Aug 26

Summary

  • A missing XMLUI How-To for converting a select widget into a radio group was added after an agent failed to find a verified pattern and had to improvise from component docs.
  • MCP log comparisons showed why the fix mattered: the failed searches returned weak, irrelevant matches around a 0.7 relevance score, while the new document ranked No. 1 at 4.2.
  • Version-pinned docs let the maintainer run a direct A/B test, and synthetic queries such as “radio buttons for a small set of options” also surfaced the new guide reliably.
  • The episode is being treated as a testable documentation bug, with XMLUI using MCP logs, application code and session logs to identify missing patterns across more apps.
  • The broader goal is to make documentation an engineered product component for both humans and agents, with gaps measured and fixed through repeatable search testing.

Insights

When software bugs are just missing words, how will technical writing evolve to control autonomous coding agents?
Could treating documentation as testable code finally stop AI assistants from hallucinating your next software project?
If AI agents are now the primary readers of software manuals, what happens to the human developers left behind?