Instagram Direct Migrates Android UI to Jetpack Compose, Cuts AI Token Costs by 33%

AI-Native Architecture for Instagram Direct

Instagram Direct's Android UI was migrated from the legacy Android View system to Jetpack Compose. This transition established an AI-native UI codebase, making it more comprehensible and maintainable for AI coding agents. The project aimed to optimize the codebase for AI interaction rather than simply using AI for piecemeal component migration.

Reduced Code Volume and AI Costs

The migration resulted in a 50% reduction in UI code volume. Furthermore, the efficiency of AI agent operations improved significantly, with token costs per AI agent session decreasing by 33%. AI agent execution time was cut by 35%, and engineer–agent exchanges were reduced by 32%.

Designing for AI Agent Behavior

A core principle of the migration was to design the architecture to guide AI agents towards correct implementations. Instagram engineers noted that AI tools often follow the path of least resistance, which can lead to bugs or technical debt. By establishing clear architectural boundaries around state management and predictable behavior, the codebase directs AI agents to produce high-quality results.

Minimizing context dependency was also a key factor. The design favors established best practices and familiar patterns over codebase-specific knowledge, aiming for more reliable AI-generated code.

Maintaining Team Productivity and Performance

The migration was executed without disrupting the wider engineering team, allowing them to continue developing new features. AI-assisted workflows were standardized to ensure consistency and adherence to shared architectural conventions. The team also prioritized preventing performance regressions, focusing on metrics like time to interact and time to fully load for screens.