TL;DR
A developer working on a GPU-aware Kubernetes dashboard decided to stop relying on AI for coding after encountering significant code quality issues. This marks a shift back to traditional hand-coding, emphasizing the limitations of current AI tools for complex projects.
A developer has publicly stated they are returning to writing code manually after experiencing critical failures in AI-generated code for a complex Kubernetes dashboard project. This decision underscores ongoing concerns about the reliability of AI-assisted coding for serious software development.
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The developer, known as shvbsle on Hacker News, detailed their experience building k10s, a GPU-aware Kubernetes dashboard, primarily with AI assistance. Over seven months, they used Claude, an AI language model, to generate features and architecture for the project, which initially sped up development significantly.
However, the developer encountered a major setback when the AI-generated code started to produce inconsistent and unstable results. After prompting the AI to develop complex components like a GPU fleet view, they found that the code became difficult to maintain and led to system failures. This culminated in the developer reading through a 1690-line file, discovering that the code had become a tangled, self-consuming structure that was impossible to fix through prompting alone.
Why It Matters
This development highlights the current limitations of AI in handling complex, interconnected software architectures. While AI can accelerate feature development, it may produce code that lacks coherence, maintainability, and robustness, especially when used without strict constraints or oversight. For developers and organizations relying on AI tools, this underscores the importance of cautious integration and thorough review processes.
Products Worth Considering
Samsung Odyssey G9 QD-OLED 49 Dual QHD Curved Gaming Monitor
LG UltraGear 32-inch Curved QHD Gaming Monitor (2560×1440, 180Hz, 1ms, 1000R, HDR10)
LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1, DisplayPort 1.4, Black
As an affiliate, we earn on qualifying purchases.
Background
The developer started using AI in late September 2025, initially achieving rapid progress on features such as resource views, log streaming, and keyboard navigation. The core goal was to build a GPU fleet view that displayed real-time metrics. The AI successfully generated initial code, but over time, the codebase grew unwieldy, with features overlapping and a single large struct becoming a bottleneck.
Previous efforts focused on speed, with the AI handling both feature creation and architectural decisions. The shift to manual review occurred after the developer identified fundamental issues in the code structure, which had been built incrementally through AI prompts without regard for overall design coherence.
“AI writes features, not architecture. The longer you let it drive without constraints, the worse the wreckage gets.”
— shvbsle
“I read the entire model.go file and was horrified. It was one big structure with no clear boundaries.”
— shvbsle
What Remains Unclear
It is not yet clear whether future AI models will better handle complex architectures or if this is a fundamental limitation. The developer has not indicated whether they plan to re-engage with AI tools under new constraints or entirely revert to manual coding.
What’s Next
The developer plans to continue developing the project manually, emphasizing careful architectural planning. They may also explore hybrid approaches that combine AI assistance with rigorous oversight and design constraints. Further updates are expected as they refine their workflow.
Key Questions
Why did the developer decide to stop using AI for coding?
The developer found that AI-generated code led to unstable, unmanageable, and self-consuming code structures, which ultimately hindered progress and project stability.
Can AI still be useful in software development?
Yes, but this experience suggests AI tools are most effective when used for specific, well-constrained tasks rather than for generating entire architectures or large codebases without oversight.
What are the risks of relying solely on AI for coding?
Risks include creating incoherent code, architectural decay, difficulty in maintenance, and potential for introducing bugs or system instability that are hard to diagnose.
Will this developer’s experience influence broader AI development in coding tools?
It may prompt developers and AI researchers to focus more on integrating architectural constraints and better oversight mechanisms in AI-assisted coding tools.
