Frontend Is Not Dead, but AI Already Holds the Shovel
A former frontend developer from Avito shares experience using AI agents in frontend development. He argues that frontend is transforming, not dying, and that AI is already useful for tasks like debugging, migrations, and generating tests. However, he warns about the growing volume of code review and the risks of 'confidently solving the wrong problem'.
Roman Mironov, a former frontend developer and current team/tech lead at Avito, explains why frontend is not dead but transforming. He notes that AI is becoming a standard tool, with 90% of developers using at least one AI tool at work, and 42% of committed code being AI-generated or AI-assisted. However, 96% of developers do not fully trust the functional correctness of generated code. He describes two cases: one where an AI agent investigated a 500 error on a micro-frontend, correctly hypothesizing that Googlebot with a mobile user agent was causing invalid content; another where an agent created 38 pull requests to remove a notes feature, a task that would have been tedious manually. He highlights the J-curve of AI adoption: initial slowdown due to learning and verification tax, then adaptation, then growth. He advises structuring prompts clearly rather than saying 'act like a senior developer', and emphasizes that engineers remain responsible for AI-generated code. He also mentions using plans in Markdown and HTML, and a multi-level config for agents, including rules and skills, but warns against overloading context.
- Abbreviations
- PR = Pull Request — запрос на слияние
- MCP = Model Context Protocol — протокол контекста модели
- JSX = JavaScript XML — расширение синтаксиса JavaScript
- CSS = Cascading Style Sheets — каскадные таблицы стилей
- A/B = A/B (split testing) — сплит-тестирование
Source: Habr — хаб ИИ —
original