In this SF Swift talk, Atharva Vaidya shares practical lessons from Portola’s use of AI in a production Apple platforms codebase. He focuses on the habits and guardrails that help teams move faster without creating slop: opinionated codebase standards, scripts for repeatable workflows, targeted skills for codebase-specific patterns, specialized agents for PRs and snapshots, and a small CLAUDE.md / AGENTS.md that points agents toward the right references. The talk also covers how the team updates skills when agents make mistakes, reviews AI-written code, manages parallel agent workflows, and uses AI to respond more quickly to bugs and user feedback while keeping quality high. Presented at SF Swift. Join future events here: https://luma.com/sfswift