Why I stopped using semantic embeddings for tool selection and switched back to BM25 [D]

I've been building agents for about a year and recently shipped one for a client running ~140 MCP-exposed tools at peak. Along the way I made the canonical mistake. I used cosine similarity over tool description embeddings to pick which tools the model could see per turn. Worked great in demos. Was actively dangerous in production. Here's the problem. In a basic semantic-ranking setup you embed the user query, embed every tool description once, and rank by cosine similarity at runtime. That works for genera