Latest / Logosofia no Dia a Dia
Measuring AI API Latency Through the Black Box
Ever felt your AI tool was sluggish, only to see a green status page? This episode dives into practical ways to measure what's happening under the hood. From Claude Code's built-in OpenTelemetry support to mitmproxy's HTTPS inspection, we explore how to capture real timing data, token counts, and rate-limit headers. You'll learn to distinguish between queuing delays, compute contention, and throttling — and finally have evidence instead of suspicion. Episode 2511 — myweirdprompts.com/n/2511 (permanent short link: myweirdprompts.com/027686)
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.