T-Mobile replaces AI agent knowledge graph with lookup table

venturebeat.com6 min read

T-Mobile discovered that loading large customer support documentation into its AI agents' context windows caused conflicting instructions and worse performance. After a knowledge graph proved too slow and difficult for agents to navigate, the company replaced it with a simple key-value database acting as a lookup table. The table maps common customer question phrasings to the correct agent path, while a second model runs alongside production to flag erroneous answers. Open article →