Semantic Routing Protocol: How AI Agents Are Starting to Talk to Each Other Directly (Not Through LLMs)

Last Updated on July 23, 2026 by Editorial Team Author(s): Rizwanhoda Originally published on Towards AI. You’ve built an AI agent that works perfectly in development. Deploy it to production with 30 different SaaS integrations and suddenly your costs are 10x higher and latency is unbearable. Here’s why, and what Semantic Routing fixes. There’s a moment every team building production AI agents hits at exactly the same place. The article argues that production agents break because they use expensive LLMs for routing/tool selection, forcing huge tool definitions into every context and causing high latency, token bloat, and hallucinated API calls. Semantic Routing addresses this by separating routing (classification) from reasoning: a fast vector classifier chooses the right intent/tool path (often in ~100ms), while the main LLM is called only for true reasoning. It explains the concept via “old vs new” flow examples, then situates the approach in timing and infrastructure changes (cost pressure, improved small routing models like vLLM Semantic Router, and emerging standardization such as IETF-backed SIRP and related ecosystem protocols like MCP and A2A). It outlines an architecture where semantic routing sits between user requests and tool selection, compares semantic routing against alternatives (LLM routing, hard-coded rules, multi-stage/hybrid methods), and highlights three practical impacts: a changed agent architecture, making multi-agent systems economically viable, and producing more predictable cost structures for pricing. Finally, it provides actionable steps (evaluate semantic routing once you have many tools, start with vLLM Semantic Router, plan for SIRP compatibility, and monitor cost baselines) and predicts semantic routing will become “table stakes” for serious production agent systems. Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor. Published via Towards AI