How we built AEO tracking for coding agents
Vercel just built the playbook for tracking how coding agents actually work—and it's nothing like monitoring LLMs.
Why it matters
As coding agents become a major traffic source (20% of prompts trigger web search), companies need observability into agent behavior. Vercel's AEO system shows how to normalize wildly different agent architectures (Claude Code, Codex, OpenCode) into a single evaluation pipeline—a pattern every infrastructure company will need.
The key facts
7 to knowCoding agents perform web searches in ~20% of prompts
Vercel built unified observability layer across Claude Code, Codex, OpenCode with different transcript formats
System uses ephemeral Linux MicroVMs with six-stage lifecycle for sandbox isolation
Four-stage normalization pipeline: transcript capture → parsing → enrichment → summary
AI Gateway routes all agent LLM calls for unified logging, rate limiting, cost tracking
Planning to open-source the AEO tracking tool for other teams
Agent recommendations embedded as working code (imports, configs, scripts) vs. prose mentions in model responses
Go to the source
Vercel Blogvercel.com
Publisher excerpt: AI has changed the way that people find information. For businesses, this means it's critical to understand how LLMs search for and summarize their web content. We're building an AI Engine Optimization (AEO) system to track how models discover, interpret, and reference Vercel and our sites. This…