Claude Code is unusable for complex engineering tasks with the Feb updates
![[MODEL] Claude Code is unusable for complex engineering tasks with the Feb updates · Issue #42796 · anthropics/claude-code](https://pub-90f0ac00d93c47daac3e1d2cdd28d496.r2.dev/articles/dc71924c247f666a79a6ac01ac2e981b.webp)
github.com
April 6, 2026
27 min read
🔥🔥🔥🔥🔥
67/100
Summary
Claude Code has become unreliable for complex engineering tasks following the February updates. Users report a significant regression in its performance, making it untrustworthy for such applications.
Key Takeaways
- The Claude Code model has regressed in its ability to perform complex engineering tasks since February, leading to significant quality issues.
- Analysis shows that the rollout of thinking content redaction correlates with a measurable decline in performance, particularly in multi-step research and code modification tasks.
- Data indicates that thinking depth has decreased by 67% from the baseline, impacting the model's effectiveness in engineering workflows.
- Users have reported that Claude now frequently ignores instructions and produces incorrect outputs, prompting some to switch to alternative providers.
Community Sentiment
Positives
- Despite recent issues, some users have successfully leveraged Claude Code for complex tasks by providing clear guidance, demonstrating the model's potential when used effectively.
- One user reported that the agents were able to research and design an app idea with minimal input, showcasing the model's capability to generate compelling solutions.
Concerns
- Users are increasingly frustrated with the model's degradation in performance, particularly when it produces useless code after initially correct reasoning, raising concerns about reliability.
- The perception that access to high-quality models is being stealthily degraded over time has led to significant distrust among users, undermining confidence in the product.