Skip to content
0:00 / 0:00

#claudecode#vibecoding#agenticai

Vibe coding complex changes in Rust

Jon Gjengset

111K subscribers • 7 videos on Hallway

YouTube
  • Recommended by 3 people: 2 on Hacker News and 1 on Lobsters.
  • Presented by Jon Gjengset, software engineer
  • Linked in the descriptionhelsing-ai/sguabaSlides on jon.thesquareplanet.com
  • May be out of date. The tools, versions or practices it shows have likely moved on.

Intermediate. Assumes working knowledge.

55K views1 year ago3 recommendations

Programming languagesAI and machine learningRustAI for codingTransformers and LLMs

Jon Gjengset live-codes complex changes in a Rust codebase using Claude Code to test the limits of agentic AI coding assistants.

Those who have followed me for a while may already know that I'm a bit of an ML Luddite — it's not that I'm opposed to the use of ML-based tools, but rather that I haven't personally found much use for them in my day-to-day work, especially when it comes to programming. My hypothesis thus far has been that this at least in part due to the nature of the work I'm doing; it often does not fit neatly into ML's strong-suit, namely pattern-replication where there are already plenty of examples for the ML to draw upon. Well, there's enough hype around vibe coding these days, and especially agentic AI coding assistants like Claude Code, that I felt like I should try it "in anger". Concretely, I have some non-trivial changes I'd like to make to the type-safe rigid body transformation crate Sguaba ( github.com/helsing-ai/sguaba ), and figured this could be a good testing ground. Sguaba doesn't have too much code, but it is sufficiently involved both in terms of the implemented logic and the use of Rust's type system that I think the ML will have its job cut out for it. This video is my attempt at disproving my own skepticism — working with Claude Code to implement some of these changes end-to-end. I also haven't used these tools myself *at all* thus far, only watched a few videos and read a few blog posts, so this is an unfiltered first-and-second impressions experience! We got through two (the easiest two) out of the four things, and then ran out of tokens 😅 But overall the experience was interesting and educational. I don't think it disproved my skepticism, though it did certainly prove that with sufficient pair programming, you can get very far. Still unclear to me that it saves time overall though, at least for specifically these kinds of tasks. If you're interested in finding out more about the sponsor of this video, Hudson River Trading, see hudsonrivertrading.com/gjengset The slides for the Sguaba presentation, you can find at jon.thesquareplanet.com/slides/sguaba/ The final PRs to Sguaba are github.com/helsing-ai/sguaba/pull/27 and github.com/helsing-ai/sguaba/pull/28 My skepticism The sguaba crate Compile-time asserting macro constructor Builder that restricts order with type-state What I wanted to do next Concluding thoughts Questions A wild sponsor appears! Live version with chat: youtube.com/live/3NGXL_8RzEI

Why it's here

3 people recommend this

What people say. Recommended for its realistic, unedited look at using agentic AI tools on a non-trivial Rust codebase.

sitkackon Hacker News • Recommends

On “AI doesn't lighten the burden of mastery”

Jon Gjengset (jonhoo) Who is famously fastidious did a stream on live coding where he did something similar in terms of control. Worth of a watch if that is a style you want to explore. link to this video I don't have the energy to do that for most things I am writing these days which are small PoC where the vibe is fine. I suspect as you do more, you will create dev guides and testing guides that can encapsulate more of that direction so you won't need to micromanage it. If you used Gemini CLI, you picked the coding agent with the worst output. So if you got something that worked to your liking, you should try Claude.

ontouchstarton Hacker News • Recommends

On “Profession by Isaac Asimov”

Ironically, when machine learning is getting “deeper & deeper”, human learning is getting more and more impatient and shallow. I have been searching “Vibe coding” videos on YouTube that are not promoting something. And I found this one and sat down and watched the whole three hours. It does take a lot of real effort. link to this video

soareschenon Lobsters • Recommends

Vibe coding complex changes in Rust

coffeerion Hacker News • Mentions

On “A staff engineer's journey with Claude Code”

This video 0 is relevant, though it actually supports your point - it shows Claude Code struggling with non-trivial tasks and needing significant hand-holding. I suspect videos meeting your criteria are rare because most AI coding demos either cherry-pick simple problems or skip the messy reality of maintaining real codebases. 0 link to this video

anuramaton Hacker News • Mentions

On “Ask HN: Share a video-series of developing non-trivial things with AI?”

take a look at the replies to link eg link and link to this video PS, what was your last setup? performance depends a lot on the domain/model/workflow1; some combinations don't work at all 1 by workflow I mean: trying to oneshot a feature VS interactive exploration VS boilerplate generation