Snap invested in its platform for over a decade, creating a robust codebase that allows non-engineers to contribute code safely. This reduces the blast radius of potential outages or performance regressions, allowing for faster iteration by breaking down traditional role barriers.
To maintain quality while increasing shipping velocity, Snap uses an AI agent named CodeBal for the first pass of all code reviews. It understands the entire codebase of millions of lines, catching complex issues humans might miss and significantly speeding up the development cycle.
Initial broad AI tool adoption led to many throwaway prototypes without business impact. Snap pivoted by defining key "jobs to be done" for each function. Now, AI development is focused on solving these specific jobs, directly tying AI usage to measurable outcomes.
Snap deployed an AI agent, Casper, that acts as a team member within tools like Slack and Jira. It listens to conversations, understands context from the entire company knowledge base and codebase, and can be invoked with a simple command to build a working prototype.
With advanced AI coding tools and a robust design system, Snap's engineers build functional prototypes so quickly that the traditional design-first workflow is inverted. Instead of waiting for Figma mock-ups, teams now build and show working code to decide which ideas win.
Snap spins off ventures like Specs (AR) to protect their startup-like nature. A mature, billion-user platform requires different operational thinking and investment strategies than a net-new product. This separation allows each entity to operate with the appropriate model for its stage.
Unlike a typical General Manager model, Snap uses a flat, functional structure where heads of product, engineering, sales, etc., all report to the CEO. This means there is no single accountable owner for major initiatives, forcing leaders to work together to get things done.