On Rails — Brian Scanlan: Building AI-First at Intercom

In this episode of On Rails, Robby is joined by Brian Scanlan, Senior Principal Engineer at Intercom, where a 15-year-old Rails monolith with millions of lines of code sits at the heart of the business.

Brian shares how Intercom's philosophy of being "technically conservative" has kept their engineering organization productive and focused on shipping product rather than managing infrastructure complexity, and on Intercom's all-in bet on Claude Code as their singular AI tool, now generating over 95% of daily code, with over 1,000 weekly users across the company including non-engineers in sales, marketing, and finance. Brian explains their approach to automated code review and PR approvals, how they built a Rails console MCP that lets Claude run production queries (with non-engineers as the top users), their layered plugin and skills architecture, and where AI still falls short in open-ended debugging, using the metaphor of commercial airline pilots who know when to disengage the autopilot.

Tools & Libraries Mentioned

• Claude Code — AI coding agent (mandated tool at Intercom) — https://docs.anthropic.com/en/docs/claude-code/overview
• Rotoscope — Shopify's tool for identifying test subsets — https://github.com/Shopify/rotoscope
• Honeycomb — Observability and tracing — https://www.honeycomb.io/
• RuboCop — Ruby linting and static analysis — https://rubocop.org/
• Snowflake — Cloud data warehouse — https://www.snowflake.com/
• Datadog — Monitoring and metrics — https://www.datadoghq.com/
• GitHub Dependabot — Automated dependency updates — https://github.com/dependabot
• Console1984 — Rails console audit trail gem — https://github.com/basecamp/console1984
•  Playwright — Browser automation (mentioned re: token consumption) — https://playwright.dev/

Books Mentioned

•  Designing Data-Intensive Applications by Martin Kleppmann 

Blog Posts Mentioned

•  Choose Boring Technology by Dan McKinley — https://mcfunley.com/choose-boring-technology

Listen to this episode (1h 51m) or subscribe to On Rails in your podcast player.