← back
How Building with AI Can Double the Throughput of Your Engineering Team — Brian Scanlan, Intercom
Takeaway
Large-scale AI productivity gains require a staffed organizational program, shared tooling, and company-specific context.
Summary
- Intercom launched a dedicated 2x initiative to double engineering throughput within a year, using code changes per R&D person as its primary measure alongside other feedback.
- Leadership made AI adoption an explicit expectation and reinforced it through repeated communication, recognition, hackathons, and immersion days.
- A full-time enablement team supports adoption across R&D rather than leaving individual engineers to establish their own workflows.
- The company standardized on Claude Code, aiming to connect it to engineering systems and teach company conventions within existing permissions and audit controls.
developer-productivityai-adoptionclaude-code
Original description
Intercom hit 2x engineering throughput in under a year. Not by prompting better. By treating Claude Code like a new hire: onboarding it to a Rails monolith built over 15 years, writing skills for every recurring task, connecting it to production systems and internal tooling, and going all in on one platform instead of letting everyone pick their favorite tool. Brian Scanlan covers what the data looks like: PR throughput doubled, 17.6% of pull requests auto-approved with SOC 2 sign-off, and the CI infrastructure collapsed under the volume. The principle behind all of it comes down to framing. Give agents problems, not tasks. He was pulled into a security incident over accidentally published Snowflake metadata, described the situation to Claude, and watched it pull the files, run the analysis, and hand back next steps in two minutes using a skill he didn't know existed. Speaker info: https://x.com/brian_scanlan / scanlanb