Yandex Browser team hits 20% dev speed gain with daily AI agent use

Yandex Browser's mobile team is running nearly all development through AI agents on a daily basis, according to a Mobile Runtime conference talk by Artur Vasilov. The team measured a roughly 20 percent speed increase in research, design, coding, and debugging over the first half of 2026, though that gain shows up in the development phase itself rather than in end-to-end time to production. Penetration is high — almost every engineer on the team now uses agent workflows for real work tasks, not just one-off chat queries.

Vasilov laid out the adoption path in layers. The first is habit formation: a shared AI chat, team ambassadors, live demos, and clean integration with Xcode, Android Studio, and the corporate environment so engineers don't get stuck on setup. Leadership matters — teams where the manager personally uses AI daily see noticeably higher adoption than those with top-down mandates alone. The second layer is infrastructure: AGENTS.md files, system prompts, Model Context Protocol hooks, and reusable skills for recurring tasks. The goal is to give the agent enough tools and context to work autonomously rather than requiring constant hand-holding.

On that foundation, heavier use cases have emerged. One is Swiftization — bulk migration of legacy Objective-C code to Swift. Another is partial automation of release duty, where an agent handles routine operations, release-state analysis, and first-pass triage of new crashes and bugs, leaving the engineer to review the output. Vasilov was explicit that without parallel automation in adjacent functions — project management, task assignment, testing — the impact on overall time to market will remain capped. The 20 percent gain is real, but it's localized to the coding loop, not the full feature pipeline.