Intercom, a leading customer messaging platform, has doubled its engineering throughput in just nine months by adopting Claude Code, an AI-driven development tool. According to Brian Scanlan, Senior Principal Engineer at Intercom, the company's success was made possible by custom skills, deep telemetry, and a culture of permission that allowed engineers to experiment with AI.
Intercom tracks skill invocations in Honeycomb and stores anonymized Claude Code sessions in S3, providing visibility into what works and what doesn't. The company also built custom dashboards that show engineers how they compare to peers, applying the same product thinking used for customer-facing features.
Intercom's partnership with Stanford researchers shows that code quality metrics are improving, not deteriorating, as a result of shipping twice as fast due to increased capacity for technical debt. Scanlan emphasizes the importance of giving permission and taking accountability in the AI era, allowing engineers to experiment without fear of repercussions.
The company has also built an Intercom CLI that can autonomously sign up for Fin, verify email addresses, and complete installation without human intervention, making the product agent-friendly and reducing the likelihood of customers building their own solutions. Scanlan predicts that all work will become agent-first in the future and sets a deadline for this shift to occur by the end of each month.
Detailed workflow walkthroughs from this episode are available on the ChatPRD AI website, providing actionable insights for companies looking to replicate Intercom's success with Claude Code.
