How Ramp Uses Claude Code to Ship 1M Lines of Code in 30 Days

Ramp has built its competitive position on the premise that financial operations should be as automated as possible. The company’s platform handles merchant-level spend restrictions, approval workflows, and expense processing for businesses that want to move faster without losing control. That same philosophy of automation extends internally: Ramp actively pursues the most advanced tools available and builds a culture of rapid adoption.

The engineering team encountered Claude Code and recognized immediately that it was different from prior AI coding tools. Rather than requiring top-down mandates, adoption spread organically—engineers who tried it independently had experiences compelling enough that they shared with colleagues without being asked. Within weeks, the tool had spread from engineering into product, design, and data teams.

Ramp built custom extensions around Claude Code to fit its specific workflows. Engineers connected it to testing frameworks via CLI for automated code-test-fix cycles. They integrated it with observability platforms—Datadog and Sentry—via Model Context Protocol servers, enabling autonomous log and error aggregation during incidents. They also connected it directly to their Snowflake data warehouse, allowing non-technical team members in product and data roles to run natural language queries without writing SQL.

The results over the first 30 days were striking. Engineers implemented more than one million lines of AI-suggested code. Weekly active usage across the engineering organization reached 50%. The new incident response tooling—built on Claude Code’s integrations with Datadog and Sentry—cut incident investigation time by 80%, reducing the time engineers spend chasing down errors during production events.

Ramp’s experience points to a broader pattern in how engineering-first companies are adopting AI coding tools: not as a productivity add-on, but as a platform layer connecting existing workflows. The spread from engineering to non-technical teams, and the 80% reduction in incident investigation time, suggests the value compounds as the tool becomes embedded in more functions across the organization.