Rewriting a Node.js Service in Go With AI Agents

Why we picked it · the editor's summary

Checkly's results daemon handles 92 million messages a day, and the team replaced it with an agent-written Go version that shipped with zero incidents. Before the rewrite they built a black-box harness from 24 hours of production configurations; CI ran it against both the legacy daemon and the new one and blocked any pull request that failed either. Traffic moved by feature flag in three cohorts, free, paid, then enterprise, with 24 to 48 hours of monitoring after each flip. The one gap: the harness routed retries to 3 queues where production uses 18, which surfaced as retry failures until the harness was corrected. The part to copy is the dual-run harness in CI, which made each cohort a flag flip you could undo.