Descriptions:
Brian Scanlan, Senior Principal Engineer at Intercom, takes the AI Engineer conference stage to share how Intercom set out to double the throughput of its engineering organization — and what it actually took to get there. Intercom, a 1,400-person B2B SaaS company that pivoted hard toward AI the week ChatGPT launched, had already shipped Finn, an AI customer support agent with over 8,000 customers and revenue approaching $100 million. The same urgency now drives internal developer productivity.
Scanlan explains that the team tried the standard tooling path — GitHub Copilot, then Cursor, then Augment — and found results marginal. The inflection came from committing fully to Claude Code as the foundation of all technical work, not just code completion. The key investment was building out Intercom-specific context: engineering skills files, guidance documents, hooks that enforce internal standards, and internal Claude plugins distributed across hundreds of developer laptops. The primary productivity metric chosen was code changes per R&D person, deliberately simple and output-oriented rather than vanity-metric-driven.
The talk walks through the principles Intercom converged on: treating agent-first workflows as the new default across debugging, testing, planning, and code production; accepting that the models are capable enough today even without further improvement; and understanding that the real bottleneck is organizational context, not model capability. Scanlan draws a parallel to the cloud transition from sysadmin to SRE — engineers move up the stack, the work becomes more impactful, and the discipline changes more than the headcount.
📺 Source: AI Engineer · Published May 15, 2026
🏷️ Format: Workflow Case Study
