Front-End Squeeze: AI Coding, Full-Stack Favoritism, and the Art of Accepting Anxiety
AI coding tools are cheapening the front-end layer in toB software, and companies picking who keeps a job are favoring back-end engineers who can carry business logic; the same calculus will reach Western firms as budgets tighten.
A 9-year front-end developer in Wuhan tells 双越 that his company laid people off heavily last December, kept trimming monthly, and is likely to cut again this winter. He expects to be caught. The company is having back-end engineers move to full-stack, while the front-end team is told to cooperate, a pattern 双越 says he hasn't seen elsewhere; other firms push both tracks together. The reasoning: this is a toB management-system business where front-end work is simple enough for AI coding, so back-end familiarity with the business carries more weight. Even the front-end team leader never advocated for the group, leaving it without leverage.
The conversation turns to AI. 双越's position: AI has hit a bottleneck and entered a stable period. Current applications split into life, office, and production; in production, AI coding and video generation are still augmentation tools that need human correction, with no autonomy. Agent work exists in the company, mostly C-end customer service and a workbuddy clone that smells like a KPI project. Still, agent development is now a mature field with ongoing demand, and enterprise adoption is early enough that it still counts as a dividend window. If an opportunity to build an agent product appears, take it.
On the personal layer, the developer is anxious about transferability, age (he's approaching 35, a steep cliff in Chinese hiring), and whether to sit for the civil-service exam. 双越 tells him anxiety and uncertainty are the baseline, that hunting for certainty inside uncertainty is a one-way trip to depression, and that the right move is to let it be: leave the lights off, stay awake, and stop trying to force sleep.
The company's preference for back-end engineers going full-stack over front-end engineers reveals that full-stack is not symmetric; business logic is the moat, and AI coding transfers simple UI work to the machine.
Calling agent development 'mature' while also saying it's in a dividend period is a useful distinction: the tooling is boring and stable, but enterprise adoption is behind, so new products can still win.
The advice to let anxiety sit uncorrected, using insomnia as the model, is a practical stance for developers in unstable job markets where no answer exists.