Post

Sam Rose
@samwho.dev

Inviting controversy with this take, which is outside of my comfort zone, but I do stand by it. You can generate good quality code with LLMs, it “just” requires a tonne of human oversight and automated guard rails. Got a week off then I’ll be building my first proper post with webernetes. Stoked.

ngrok@ngrok.com

Our own @samwho.dev ported Kubernetes to the browser. Like a real flippin' cluster with lifecycles, DNS, and a simulated network. It's ~100k lines of TypeScript, almost all written by LLMs, but with every line reviewed by hand to keep it slop-free ↓ ngrok.com/blog/i-porte...

ngrok.comI ported Kubernetes to the browser | ngrok blogAlmost 100,000 lines of LLM-generated code in 2 months, and none of it is slop.
hayleigh
@hayleigh.dev

pre-llms would you have trusted an engineer that said they had thoroughly reviewed 100k loc in a month? ive never worked anywhere big-techish that pushes code at that scale so i cant imagine doing it without a healthy dose of LGTM

Sam Rose
@samwho.dev

It’s about 1600 lines a day, and most of the review was side-by-side checking that the TS version matched what the Go version was doing, which is more tedious reviewing but less thinking required. I think it’s reasonable, yeah.

0 reposts1 like

Sam Rose
@samwho.dev

Actually I counted weekends there. It’s more like ~2100/day. Still well within the realms of reason, especially when it’s such a mechanical review process. Definitely wasn’t fun, mind 😅

Sam Rose
@samwho.dev

I’ll add to this and say if the figure was 100,000 in one month I would be a lot more doubtful. I know I worked _hard_ over those two months. It was two months purely because my reviewing code was the bottleneck.

Sam Rose
@samwho.dev

I also know my review wasn’t perfect. A couple times I’d find mistakes in files I’d already combed through and kick myself for missing it. I do expect to find bugs as I begin using this lib in earnest.