Brian Scanlan is a senior principal engineer at Intercom leading developer-infrastructure work and helping redesign software development around AI coding agents. His approach combines the operational discipline of platform engineering with an unusually concrete ambition: double engineering throughput without doubling the team.
From infrastructure to engineering leadership
Scanlan worked at HEAnet and Amazon before joining Intercom. He began there as an engineer, became an engineering manager, and subsequently managed other managers—a progression he described in his account of moving from maker to manager. His responsibilities have encompassed reliability, security, observability, infrastructure costs, and the developer experience around Intercom’s large Ruby on Rails application.
2X engineering-throughput initiative.Intercom measured code changes per research-and-development employee, created a dedicated adoption team, trained engineers, and made AI fluency an organizational expectation. Scanlan reported that pull-request throughput doubled in less than a year while acknowledging the limitations of productivity metrics. His AI Engineer Europe talk makes the case for treating adoption as an engineering-wide operating change.
Durable, testable agent skills. Standardizing on Claude Code gave teams a shared platform for encoding Rails conventions, architectural decisions, security policies, testing standards, and internal procedures. Scanlan prioritizes reusable organizational knowledge over bespoke orchestration, including a skill for repairing flaky Rails tests developed through iterative feedback.
Problem-first agent workflows. Engineers should describe the outcome they need and let agents locate appropriate tools and guidance. Scanlan applied this approach to a security incident, where an agent found the relevant internal policy, assessed exposed material, and proposed next steps.
Evaluated automatic code review. Narrowly scoped pull requests can earn automatic approval through historical backtesting, human-labeled evaluations, multiple reviewing agents, and auditable controls. Instrumentation of skill usage and agent sessions provides feedback for improving both the automation and the surrounding engineering environment.
For Scanlan, agent-first development continues the shift he experienced moving from Unix systems administration into cloud infrastructure: engineers gain leverage by defining problems, encoding institutional expertise, and building systems that let capable automation operate safely.
