[zh3]

I'll bite.

Here's my realtime Bluetooth heart rate monitor for linux, with text output and web interface.

   https://github.com/lowrescoder/BlueHeart

This was 100% written by Claude Code, my input was limited to mostly accepting Claude suggestions except a couple of cases where I could make suggestions to speed up development (skipping some tests I knew would work).

Particularly interesting because I didn't expect this to work, let along not to write any code. Note that I limited it to pure C with limited dependencies; initial prompt was just to get text output ("Heart Rate 76bpm"), when it got to that point I told Claude to add a web interface followed by creating a realtime graph to show the interface in use.

Every file is claude generated. AMA.

edit: this was particularly interesting as it had to test against the HRM sensor I was wearing during development, and to cope with bluetooth devices appearing and disappearing all the time. It took about a day for the whole thing and cost around $25.

further edit: I am by no means an expert with Claude (haven't even got to making a claude.md file); the one real objective here was to get a working example of using dBus to talk to blueZ in C, something I've failed at (more than once) before.

[embedding-shape]

It's a good demonstration of when agents still don't get everything right when you place things into Markdown documentation. You have to be really valiant and verify everything from top to bottom, if you want to control how things are implemented to that degree, otherwise the agent will still take shortcuts where they can.

In https://github.com/lowrescoder/BlueHeart/blob/68ab2387a0c44e... for example, it doesn't actually do SSE at all, instead it queues up a complete HTTP response each time, returns once and then closes the stream, so basically a normal HTTP endpoint, "labeled" as a SSE one. SSE is mentioned a bunch of times in the docs, and the files/types/functions are labeled as such, but that doesn't seem to be what's going on internally, from what I could understand. Happy to stand corrected though!

[zh3]

Yes, I haven't even read most of the files, just threw it up there as an example for the OP (I too am tired of the lack of examples, so stepped up to the plate on this one).

As a personal bit of development last weekend. I can see inconsistences myself, some of which result from scope creep during development (starting with the idea of a text-only app and then grafting on the web side) - it literally only started because I wanted a working example of bluetooth and dBus in C, the rest of it just joined the ride.

As for the SSE, no expert on that myself, however if you watch the messages in the browser console it appears to push updates with sporadic notes about using polling instead.

[embedding-shape]

> Yes, I haven't even read most of the files, just threw it up there as an example for the OP (I too am tired of the lack of examples, so stepped up to the plate on this one).

Right, kind of like an LLM skimming and missing the core points :)

OP didn't ask for "Anything you've vibe-coded" but explicitly asked for code written with LLMs that is high quality and structurally sound, and "creates more value than it creates technical debt". That's why I felt like reviewing the code in the first place, and why I gave the feedback.

I understand now that maybe it felt like my impromptu code review came out of nowhere, but I thought you were actually trying to give OP a accurate sample, so sorry if it felt like it came out of nowhere :)

[zh3]

NP, and the exact definition of vibe-coding is, I think, yet to be determined. This wasn't a yolo, it was read all the prompts and generally accept them. Overall I'd say the code and web page are at least of a quality I've seen in many commercial settings; the code itself looks reasonable and if I was to do anything to it for a real 'release', I'd update the documentation which has suffered due to the extensive scope creep during implementation.

[embedding-shape]

> the exact definition of vibe-coding is, I think, yet to be determined

Huh? No, that's been established since Karpathy coined the term; you don't review the code, only use the agent and don't care about how it was done, just about the results.

The actual interesting stuff is how to use LLMs together with a human, to build high quality code. More "augmenting the human intellect" rather than "autonomous robots building for you".

Overall I'd say if someone handed you a specification that named SSE specifically, you created files with SSE in the name, and the implementation talks about doing SSE, yet it doesn't actually do SSE in the end, it's pretty much on par with code in commercial settings, yeah :) But maybe our bar should be slightly above the ground at least? :)

[maerch]

> Huh? No, that's been established since Karpathy coined the term; you don't review the code, only use the agent and don't care about how it was done, just about the results.

However, nowadays it is used as a synonym for everything that is somehow generated by an LLM. Regardless of whether it is a spec-driven, carefully reviewed and iterative piece of software or some yolo-style one-prompter with no idea how it was done.

[clbrmbr]

Yes, an understanding of sockets and timing of interprocess communication & networking seems to be a weak point of current models.

[kqr]

Have you reviewed the code? What were the problems with it? Where did it do things better than you'd expect of humans? Have you compared the effort of making changes to it to the effort required for similar, human-written software?

I don't think anyone says it's not possible to get the LLM to write code. The problems OP has with them is that the code they write starts out good but then quickly devolves when the LLMs get stuck in the weird ruts they have.

[zh3]

Far short of a proper review, however I have scanned the code. Bear in mind this was a purely personal project, never intended to see the light of day and initially just done to create a small but operable chunk of dbus/blueZ glue code for another project.

I have no doubt that a C developer with sufficient knowledge of dBus, bluetooth, the HRM profile and linux could have written the C code in a day. Adding the HTTP server again would be easy if the developer also had experience of that (n.b. there was a minor compiler error when I tried it on another system due to a slightly different version of libmicrohttpd). Adding the API would be straighforward (but tedious) and similiarly the web page (the web page was an one-shot after Claude wrote the API, vis. "Create a web page to display a real time plot with history using the API").

So overall I'd answer that that human developers would could have pulled that off in a day are few and far between (and likely to cost a lot more than $25 plus a day of my time).

And do I think the code is good enough? Yes, more than good enough. I could take it and run with it, against that because it ended up 100% AI-generated I feel a bit like leaving it as a monument to "pure AI".

After all, I never intended to release it - it was this thread that made my throw it up on Github as an example for the OP.

[59nadir]

Thank you for actually posting an example that people can look at; I think most other responders misunderstood the post as asking for more pointless anecdotes filled with superlatives and "trust me bro" sentiments.

[bobmcnamara]

Pretty neat!

Is there a name for the UI style of the web server page? I've noticed several web apps have a similar style to that.

[zh3]

I didn't ask for a style, it's just what Claude came up with by default. Here's what happened just now when I asked it:-

   What was the inspiration for the CSS styling of this web page?                                                                    

  ● Looking at the CSS styling in live_plot.html, the inspiration appears to be Tailwind CSS and modern minimalist design trends.

[embedding-shape]

Just as a heads up, LLMs doesn't actually understand why they do what they do, you asking about it will make them reason about why it happened, but it's not the "motivation", it's essentially guesses with no anchoring to reality.

Just thought I'd clarify as I've seen prompts like this and people thinking this is the actual motivation from the "inside the LLM" or whatever, which is a bit far away from the truth.

[zh3]

Fair enough, I did ask for "inspiration" through rather than "motivation" - mainly because I recall a comment on here a few days ago that LLMs are carefully trained to never reveal where the training material came from. So the prompt was aimed at working around that.

[embedding-shape]

Yeah, inspiration, motivation, justification etc are synonyms in this case, the point I was trying to make was something like "LLMs don't know why they do what they do", and asking for them to provide it, will make them come up with it on the spot afterwards, not actually share what the inspiration/motivation/justification was at the time the tokens were sampled.