Case study
Ramp's 80% Incident Reduction
Investigating prod incidents got 80% faster. The pattern that worked at Ramp generalizes to any oncall rotation.
Ramp
80% faster incident investigation
Ramp wired Claude Code into their incident response loop. When an alert fires, the agent pulls the relevant logs, traces, recent deploys, and PRs, then summarizes a likely cause and a suggested next step. Engineers said the agent didn't replace their judgment — it eliminated the 'gather context' phase that used to eat the first 20 minutes of every incident.
The Ramp pattern is generic: incidents have a predictable context-gathering phase (logs, traces, recent changes) followed by a judgment phase (what's the likely cause, what do we do?). Agents are exceptional at the first phase and bad at the second. Splitting the work along that line lets you keep humans where they're needed and offload the rest.
What to copy: build a /incident slash command that takes an alert ID and runs the full context-gather automatically. Your oncall engineer wakes up at 3am to a pre-summarized investigation, not a blank page.