Frontend · Backend · Artificial Intelligence

Laid Off, Paid in Full: A Week of Chinese Dev Life in Severance, AI Bugs, and Performance Theater

By 掘金一周 · Jul 8, 2026

Read original on juejin.cn ↗ Google Translate ↗ Alt translation

The digest is a raw pulse-check on Chinese tech employment in mid-2026. Layoffs are happening but severance norms appear intact; AI coding tools are mainstream enough to cause real workflow damage; and performance-review theater is so normalized that workers openly game it. For anyone tracking the global developer experience, the emotional register here — resignation without rage — is the story.

Summary

One developer walked out of a layoff with full severance and no drama — just reflection. Another watched an AI agent overwrite a day's worth of UI polish because the work hadn't been committed to Git. A third discovered that kimi code fixed in one run a bug that claude code couldn't solve in half a day, despite both running the same underlying model.

On the performance-review front, the cynicism is open. One engineer handles their own development, test-case writing, execution, bug fixes, and test reports — calling it worse than slavery. Another learned from veteran colleagues to always self-rate at 100%, because no one else will. A third sat through a meeting where the leader calculated that working three weeknights until 8:30 PM plus every Saturday would net roughly an extra 1,000 yuan a month.

Elsewhere, a first paycheck of 4,700 yuan in Beijing brought a programmer to tears on the bus, and a newlywed thanked a snack brand for the wedding gift of spicy strips.

Takeaways

— Layoffs are occurring at Chinese tech firms, but at least one company paid full severance with no pushback for roughly seven or eight employees.

— An AI coding agent overwrote a developer's day-long UI work because the changes had not been committed to Git beforehand.

— kimi code fixed a bug in a single run that claude code could not resolve in half a day, even though both tools used the same underlying model (kimicode 2.7).

— Cursor's available model list has shrunk; one user reports only a few options remain, with the default being the self-developed composer 2.5.

— One developer handles the entire QA cycle alone — writing test cases, executing them, fixing bugs, and producing the test report.

— Veteran employees in a performance-review system always self-rate at 100%, reasoning that no one else will advocate for them.

— A team lead calculated that working overtime three weeknights until 8:30 PM plus every Saturday would add roughly 1,000 yuan to a monthly paycheck.

— A Beijing-based developer's first monthly paycheck was 4,700 yuan, which brought them to tears on the bus ride home.

Conclusions

AI coding tools are now routine enough that losing uncommitted work to an agent is a relatable, shared frustration — not an edge case. The tooling has outpaced the discipline around version control that would protect against it.

The kimi code vs. claude code comparison is notable because both ran the same underlying model, yet produced different outcomes. This points to differences in prompt engineering, context handling, or toolchain integration mattering more than the model itself.

Self-rating at 100% in performance reviews is treated as obvious survival strategy, not dishonesty. When the system is perceived as zero-sum, gaming it becomes the rational norm.

The overtime math — 1,000 yuan for roughly 16 extra hours a week — prices a developer's evening and weekend time at about 15 yuan per hour, a figure the leader presented as an incentive rather than a warning.

Source: juejin.cn ↗ Google Translate ↗ Backup ↗