跪拜 Guibai
← All articles
Frontend · JavaScript · Programmers

AI Didn't Kill Frontend — It Killed Pixel-Pushing

By ErpanOmer ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

AI code generation is collapsing the bottom of the frontend market, but it also exposes a durable premium: engineers who design data protocols, harden polling against network collapse, and make build-vs-buy calls that protect margin. The career risk is not AI itself — it is staying in the pixel-pushing tier that AI now does for free.

Summary

Large models now generate entire admin panels from a sketch in minutes, which vaporizes the value of hand-coding layouts and translating Swagger docs. The frontend roles getting cut are the ones still doing that mechanical work. Surviving the shift means moving into the layers AI cannot touch: network resilience under flaky conditions, memory-safe polling, and intercepting dirty backend data through a BFF or edge middleware so the browser receives only clean, pre-shaped payloads. The highest-leverage skill is knowing when not to build — embedding a third-party SaaS or a crude iframe to validate a business idea costs nothing and earns more trust from a CTO than a month spent on perfect generic components. Five years in, the salary ceiling is set by sensitivity to extreme environments, control over team throughput, and reliable delivery, not by how fast someone can write a `map` loop.

Takeaways
AI tools such as Codex and Claude Code can produce complete, styled business components from a design link in under a minute, which eliminates the need for manual layout translation.
Mechanical frontend work — pixel-matching Figma, transcribing Swagger types, writing boilerplate `map` loops — is the exact layer being automated and cut.
Controlling the data shape matters more than consuming endpoints: inserting a BFF or a Cloudflare Worker to clean, aggregate, and cache data before it hits the browser removes entire classes of frontend complexity.
Standard polling code from AI ignores tab visibility, in-flight request overlap, rate limiting, and memory leaks; a senior engineer adds visibility gates, circuit breakers, backoff, and abort cleanup to keep the system alive under real-world stress.
Not building is often the best architecture — embedding a third-party SaaS or an iframe to test a feature costs zero engineering cycles and is valued more by leadership than a polished in-house component that delays learning.
The salary ceiling after five years hinges on production resilience, data-flow ownership, business judgement, and delivery reliability, not on framework speed.
Conclusions

The argument reframes AI not as a job destroyer but as a class divider: it automates the bottom tier of frontend work and simultaneously raises the premium on defensive engineering and architectural decision-making.

Polling resilience is treated as a litmus test — the example code is deliberately simple to show that even a trivial feature separates a safe production system from a brittle one when network conditions degrade.

The piece treats 'not writing code' as a senior business instinct rather than laziness, which inverts the typical engineering pride in building from scratch and aligns with how CTOs actually measure value: speed of validated learning over code elegance.

Concepts & terms
BFF (Backend For Frontend)
A dedicated server layer that sits between the frontend and downstream microservices, reshaping, aggregating, and caching data so the client receives exactly the payload it needs without complex client-side data wrangling.
Circuit breaker / backoff
A resilience pattern where a client stops sending requests when the server signals overload (e.g., HTTP 429) and waits an increasing delay before retrying, preventing cascading failure.
Edge middleware (Cloudflare Workers)
Code that runs on a CDN edge node close to the user, capable of intercepting and transforming HTTP requests and responses before they reach the origin server or the browser.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗