I’m going back to writing code by hand

A developer has declared they are returning to writing code by hand after a long experiment with AI-assisted development led to a problematic, unmanageable codebase.

The developer, known on Hacker News for their work on a GPU-aware Kubernetes dashboard called k10s, shared that after seven months of vibe-coding with Claude, their AI assistant, they encountered critical issues that forced a reevaluation of their approach.

They detailed how AI-generated features initially accelerated development but eventually resulted in a complex, tangled codebase. The project’s core component, a large ‘Model’ struct, grew unwieldy, leading to bugs and a breakdown in functionality, such as stale data displays and broken live updates. The developer admitted that their reliance on AI to build features without considering overall architecture caused the project to collapse.

Why It Matters

This development underscores ongoing concerns about AI-assisted coding, especially regarding maintaining code quality, architecture, and long-term stability. It highlights the risks of over-relying on AI for feature development without proper architectural oversight, which can lead to technical debt and project failure.

Background

The developer started the project in late September 2025, leveraging AI to rapidly add features such as resource views, live updates, and GPU monitoring for Kubernetes clusters. Initial progress was promising, with AI producing functional code in a matter of weekends. However, as the project grew, the AI-generated code became increasingly complex and difficult to manage, culminating in a critical failure when core features stopped working and the code structure became unmanageable.

“AI writes features, not architecture. The longer you let it drive without constraints, the worse the wreckage gets.”

— the developer

“I learned over these 7 months that what AI produces can be more damaging than helpful if not carefully managed.”

— the developer

What Remains Unclear

It remains unclear whether the developer’s return to manual coding will fully resolve the issues or if similar problems might recur in future projects. The broader implications for AI-assisted development practices are still being evaluated.

What’s Next

The developer plans to rebuild the project manually, focusing on better architecture and modular design. They also intend to document lessons learned to guide others in avoiding similar pitfalls. Monitoring will continue to see if manual development yields more stable results.

Key Questions

Why did the developer decide to stop using AI for coding?

The developer found that AI-generated code led to a tangled, unmanageable codebase that caused features to break and the project to collapse. They concluded that AI writes features but does not understand or preserve overall architecture.

What specific problems did the AI-generated code cause?

The code became overly complex, with a large ‘Model’ struct that grew unwieldy, leading to bugs, stale data, and broken live updates. The project’s core features failed due to lack of architectural oversight.

Is this a common issue with AI-assisted coding?

While AI can accelerate feature development, this case illustrates that without careful management, it can produce problematic code structures that are hard to maintain, a concern shared by some experts in the field.

Will the developer continue using AI in the future?

The developer has not ruled out future AI use but emphasizes the need for strict architectural constraints and oversight to prevent similar issues.