Yogiyo Product Engineer Pilot: Context and Verification, Not Code Generation, Are the Bottlenecks

• Code generation speed has increased, but the time to deployment has not been shortened; the bottlenecks lie in context, verification, and processes
• The points where AI cannot provide answers using only documentation and the codebase represent the organization's tacit knowledge, which must be documented for AI to reference
• A PE pilot, where engineers take full responsibility from problem definition to deployment across functional boundaries, was conducted in Q2 2026
• The mobile team is expanding its scope of work to include Android, iOS, and BFF servers starting from Q3 to address team-level bottlenecks

PE Pilot Execution Principles

• Set aside platform expertise and define business problems from a blank slate
• Focus on verifying whether the intent and quality of AI-implemented results align, rather than writing code directly
• Ask AI first, based on documentation and the codebase, before asking colleagues

Identified Bottlenecks and Costs

• Logger design and verification: Writing code incurred less cost than checking taxonomy alignment with the data team and verifying event occurrence on actual devices
• Code review waiting: Significant waiting time and response costs for PR approvals and merges across the BE, Android, and iOS platforms
• Deployment cycle: A lead time exists where missing the bi-weekly deployment calendar requires waiting another two weeks
• Lack of context: Undocumented dependencies between microservices and API version history are difficult for AI to grasp