Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Scaling long-running autonomous coding

cursor.com

Post reply on HN
Print view View original

121–130 of 203 posts

Re: Scaling long-running autonomous coding

#121

Post by underdeserver »

physicsguy wrote: post #118

I have been trying Claude Code a lot this week. Two projects: * A small statically generated Hugo website but with some clever linking/taxonomy stuff. This was a fairly self-contained project that is now 'finished' but wouldn't hvae taken me more than a few days to code up from scratch. * A scientific simulation package, to try and do a clean refresh of an existing one which i can point at for implementation details…

On Twitter people are saying GPT-5.2 is better. That's also what Cursor used in their testing. Maybe try it?

Re: Scaling long-running autonomous coding

#122

Post by underdeserver »

embedding-shape wrote: post #109
snek_case wrote: post #68

Earlier quoted context omitted.

I found the codebase very hard to navigate. Hundreds (over a thousand?) tiny files with less than 200 lines of code, in deeply nested subdirectories. I wanted to find where the JavaScript engine was, and where the DOM implementation was located, and I couldn't easily find it, even using the GitHub search feature. I'm not exactly sure what this browser implements and how. Even their README is kind of crappy. Ideally y…

> It's impressive that they got such a big project to be built by agents and to compile But that's the thing, it doesn't compile, has a ton of errors, CI seems broken since long... What exactly is supposed to impressive here, that it managed to generate a bunch of code that doesn't even compile? What in the holy hackers is this even about? Am I missing something obvious here? How is this news?

Looks like it doesn't compile for at least one other guy (I myself haven't tried): https://github.com/wilsonzlin/fastrender/issues/98

Yeah, answers need to be given.

Re: Scaling long-running autonomous coding

#123

Post by physicsguy »

underdeserver wrote: post #121
physicsguy wrote: post #118

I have been trying Claude Code a lot this week. Two projects: * A small statically generated Hugo website but with some clever linking/taxonomy stuff. This was a fairly self-contained project that is now 'finished' but wouldn't hvae taken me more than a few days to code up from scratch. * A scientific simulation package, to try and do a clean refresh of an existing one which i can point at for implementation details…

On Twitter people are saying GPT-5.2 is better. That's also what Cursor used in their testing. Maybe try it?

I have Web access for ChatGPT through work, but not API access annoyingly.

Re: Scaling long-running autonomous coding

#124

Post by heliumtera »

observationist wrote: post #6
jphoward wrote: post #2

The browser it built, obviously the context window of the entire project is huge. They mention loads of parallel agents in the blog post, so I guess each agent is given a module to work on, and some tests? And then a 'manager' agent plugs this in without reading the code? Otherwise I can't see how, even with ChatGPT 5.2/Gemini 3, you could do this otherwise? In retrospect it seems an obvious approach and akin to how…

Get a good "project manager" agents.md and it changes the whole approach of vibe coding. For a professional environment, with each person given a little domain, arranged in the usual hierarchy of your coding team, truly amazing things can get done. Presumably the security and validation of code still needs work, I haven't read anything that indicates those are solved yet, so people still need to read and understand t…

I tried this approach yesterday and I`m loving our daily standup with the agents. Looking forward to our retro and health-checks rituals

Re: Scaling long-running autonomous coding

#125

Post by LiamPowell »

qingcharles wrote: post #81
tired_and_awake wrote: post #28

The moment all code is interacted with through agents I cease to care about code quality. The only thing that matters is the quality of the product, cost of maintenance etc. exactly the thing we measure software development orgs against. It could be handy to have these projects deployed to demonstrate their utility and efficacy? Looking at PRs of agents feels a wrong headed, like who cares if agents code is hard to r…

We don't read the binary output of our C compilers because we trust it to be correct almost every time. ("It's a compiler bug" is more of a joke than a real issue) If AI could reach the point where we actually trusted the output, then we might stop checking it.

> "It's a compiler bug" is more of a joke than a real issue

It's a very real issue, people just seem to assume their code is wrong rather than the compiler. I've personally reported 12 GCC bugs over the last 2 years and there's 1239 open wrong-code bugs currently.

Here's an example of a simple one in the C frontend that has existed since GCC 4.7: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105180

Re: Scaling long-running autonomous coding

#126

Post by missingdays »

sashank_1509 wrote: post #4

Can a browser expert please go through the code the agent wrote (skim it), and let us know how it is. Is it comparable to ladybird, or Servo, can it ever reach that capability soon?

You can start by trying to compile the project (spoiler: you can't)

Re: Scaling long-running autonomous coding

#128

Post by Bishonen88 »

physicsguy wrote: post #118

I have been trying Claude Code a lot this week. Two projects: * A small statically generated Hugo website but with some clever linking/taxonomy stuff. This was a fairly self-contained project that is now 'finished' but wouldn't hvae taken me more than a few days to code up from scratch. * A scientific simulation package, to try and do a clean refresh of an existing one which i can point at for implementation details…

Similar experience here.

Did sign up for Claude Code myself this week, too, given the $10/month promo. I have experience with AI by using AWS Kiro at work and directly prompting Claude Opus for convos. After just 2 days and ~5-6 vibe coding sessions in total I got a working Life-OS-App created for my needs.

- Clone of Todoist with the features that I actually use/want. Projects, Tags, due dates, quick adding with a todoist like text-aware input (e.g. !p1, Today etc.)

- A fantastical like calendar. Again, 80% of the features I used from Fantastical

- A Habit Tracker

- A Goal Tracker (Quarterly / Yearly)

- A dashboard page showing todays summary with single click edit/complete marking

- User authentication and sharing of various features (e.g. tasks)

- Docker deployment which will eventually run on my NAS

I'm going to add a few more things and cancel quite a few subscriptions. It one-shots all tasks within minutes. It's wild. I can code but didn't bother looking at the code myself, because ... why.

Even though do not earn US Tech money, am tempted to buy the max subscription for a month or two although the price is still hard to swallow.

Claude and vibe coding is wild. If I can clone todoist within a few vibe coding sessions and then implement any additional/new feature I want within minutes instead proposing, praying and then waiting for months, why would I pay $$$...

Re: Scaling long-running autonomous coding

#129

Post by machiaweliczny »

viraptor wrote: post #101
hahahahhaah wrote: post #90

Earlier quoted context omitted.

Web browser should be easy as source exists. Fix all SVG bugs in my browser tho...

There are 3.5 serious open codebases of web browsers currently. Only two are full featured. It's not nothing, but it's very far from "source exists so it's easy to copy what they do".

But detailed specs exists for both HTML and JS and tests also exists and unlimited amount of test data. You can just try running webpage or program and also have reference implementations - it's much easier for agents to understand that. Also HTML they know super well from scraping whole internet but still impressive.

Re: Scaling long-running autonomous coding

#130

Post by ramon156 »

> A long-running agent made video rendering 25x faster with an efficient Rust version.

Which is not an optimization. This is coming from a Rust dev; Rewriting it in Rust is not the optimization.

Also, I do not believe they actually reviewed the SolidJS->React PR. This PR is incredibly unrealistic and should've been done with either stacker PRs or incremental non-breaking changes.

None of this feels organic, can we stop pretending it is?

To continue the pessimistic tone, none of the writing went in-depth. I did not gain any knowledge, just a marketing post.

Post reply on HN