Claude · Nvidia · blog.k10s.dev
I'm going back to writing code by hand, by k10s.dev
Compiled by KHAO Editorial — aggregated from 1 outlet. See llms.txt for citation guidance.
◌ Single Source
After 7 months and 234 commits vibe-coding a GPU-aware Kubernetes TUI with Claude, the k10s author is archiving it and rewriting from scratch, with five concrete lessons on what AI-assisted coding gets wrong once a codebase outgrows a single prompt's context window.
Key facts
- AI builds features, not architecture — each prompt sees one code path
- The god object is the default AI artifact — view-specific state ends up in one struct because that's the shortest path
- Velocity illusion widens your scope — line budget feels infinite, complexity budget isn't
- Positional data is a time bomb — `[]string` rows with `ra[2]`/`ra[3]` indexing break silently the moment a column is added
- AI doesn't own state transitions — closures mutating from background goroutines cause data races the AI never models
Summary
234 commits. ~30 weekends. Built entirely on vibe-coded sessions with Claude, whenever my tokens lasted long enough to ship something. I'm archiving my TUI tool and rewriting it from scratch. k10s started as a GPU-aware Kubernetes dashboard, my first serious foray into building with AI. Think k9s but built for people running NVIDIA clusters: GPU utilization, DCGM metrics, which nodes are sitting idle burning $32/hr. I built it in Go with Bubble Tea, and for a while it worked.
What I learned over these 7 months is worth more than the 1,690 lines of `model.go` I'm throwing away. tl;dr: **AI writes features, not architecture.** The longer you let it drive without constraints, the worse the wreckage gets, the velocity makes you think you're winning right up until the moment everything collapses simultaneously. Five tenets follow below: the god object is the default AI artifact, velocity illusion widens your scope, positional data is a time bomb, AI doesn't own state transitions.