Testing in production

Refactoring at Scale Done Right 🏗️ — with Maude Lemaire

Topic 9 of 9

Published Jul 24, 2025 · Clip 58:01–1:04:37 (6:36)

Maude embraces testing in production as a spicy take, noting she spent her last four Slack years sending synthetic traffic to production 24/7 because truly replicating production in staging is prohibitively expensive and never accurate—citing the New Year phenomenon where Slack's servers scale down and then can't scale back fast enough without preemptive warming. She argues new companies should especially test in production since risk is lower, and even big tech runs synthetic load in production-like systems. She wishes Slack had invested in something like Meta's Kraken system, which artificially restricts traffic to subsets of edges to stress-test nodes and can be turned off once graphs go red.