Agents will make your telemetry explode. You are not ready. – shipping bytes

AI coding agents like Claude Code are already driving 30% of Vercel deployments, up 1000% in six months. This explosive growth will overwhelm observability stacks built for human-paced deployments. The core problem is twofold: volume and context loss. The three-pillars model (metrics, logs, traces in silos) destroys context at write time, making it impossible to trace why something broke. OpenTelemetry with proper W3C trace context propagation can bridge these signals, but most teams stop at basic auto-instrumentation. Agent-initiated deployments add a deeper problem: the agent's reasoning and tool calls are invisible unless explicitly instrumented. Teams need to treat agent traces as first-class telemetry, designing structured logging of tool calls and context propagation from agent decision points to production incidents before agents ship anything.