21–30 of 42 posts
This is so poorly written. What is "Ralph"? What is its purpose? How does it work? A single sentence at the top would help. The writer imagines that the reader cares enough to have followed their entire journey, or to decode this enormously distended pile of words. More generally, I've noticed that people who spend a lot of time interacting with LLMs sometimes develop a distinct brain-fried tone when they write or ta…
Has anyone used this technique with other LLMs that are good at coding but not so expensive: for example Qwen 3 Coder?
I think the key to it is having lots of smaller tasks with fresh context each loop. Ralph loop run starts, it picks the most important task, completes it, and ends its loop. Then the next ralph run starts with new context, grabs the most important task, and the loops continue. I have not tried this method yet.
This is so poorly written. What is "Ralph"? What is its purpose? How does it work? A single sentence at the top would help. The writer imagines that the reader cares enough to have followed their entire journey, or to decode this enormously distended pile of words. More generally, I've noticed that people who spend a lot of time interacting with LLMs sometimes develop a distinct brain-fried tone when they write or ta…
Please don't post shallow dismissals of other people's work (this is in the site guidelines: https://news.ycombinator.com/newsguidelines.html ) and especially please don't cross into personal attack.
Earlier quoted context omitted.
I don’t think anyone serious would recommend it for serious production systems. I respect the Ralph technique as a fascinating learning exercise in understanding llm context windows and how to squeeze more performance (read: quality) from today’s models Even if in the absolute the ceiling remains low, it’s interesting the degree to which good context engineering raises it
How is it a “fascinating learning exercise” when the intention is to run the model in a closed loop with zero transparency. Running a black box in a black box to learn? What signals are you even listening to to determine whether your context engineering is good or whether the quality has improved aside from a brief glimpse at the final product. So essentially every time I want to test a prompt I waste $100 on Claude…
Earlier quoted context omitted.
Please don't post shallow dismissals of other people's work (this is in the site guidelines: https://news.ycombinator.com/newsguidelines.html ) and especially please don't cross into personal attack.
How is it shallow? The commenter asked three questions. That shows that they read the article and are reacting to it. Shallow would be something like, "More AI slop."
So their comment is really a vehicle for them to deliver an insult and doesn't represent significant engagement with the material or a thoughtful digression that could foster curious conversation.
Note that that doesn't mean it's a good article or that Ralph is a good idea.
Earlier quoted context omitted.
It's explained under the July 2025 heading with link to the blog post where it was first shared. The key bit is right under that though. Ralph is literally just this: while :; do cat PROMPT.md | npx --yes @sourcegraph/amp ; done
Surely that would be better written as cat PROMPT.md | cat | npx --yes @sourcegraph/amp
If you'll code a demo MVP, one-off idea, etc, alright, go ahead, have your fun and waste your tokens.
However I'll be happy when the (forced) hype fades off as people realise that there's nothing novel, insightful or even well-defined behind Ralph.
Loops have always been possible. Coordination frameworks (for tracking TODOs, spawning agents, supervising completion, etc) too, and would be better embodied as a program instead of as an ad-hoc prompt.
I forked the anthropic Ralph Wiggum plugin: https://github.com/jes5199/chief-wiggum there’s some debate about whether this is in the spirit of the _original_ Ralph, because it keeps too much context history around. But in practice Claude Code compactions are so low-quality that it’s basically the same as clearing the history every few turns I’ve had good luck giving it goals like “keep working until the integration t…
Ralph is, very literally, vibe coding with extra steps. If you'll code a demo MVP, one-off idea, etc, alright, go ahead, have your fun and waste your tokens. However I'll be happy when the (forced) hype fades off as people realise that there's nothing novel, insightful or even well-defined behind Ralph. Loops have always been possible. Coordination frameworks (for tracking TODOs, spawning agents, supervising completi…
With YOLO on full-auto, you can give a wrapping rule/prompt that says more or less: "Given what I asked you to do as indicated in the TODO.md file, keep going until you are done, expanding and checking off the items, no matter what that means -- fix bugs, check work, expand the TODO. You are to complete the entire project correctly and fully yourself by looping and filling in what is missing or could be improved, until you find it is all completely done. Do not ask me anything, just do it with good judgement and iterating."
Which is simultaneously:
1. an effective way to spend tokens prodigiously
2. an excellent way to to get something working 90% of the way there with minimal effort, if you already set it up for success and the anticipatable outcomes are within acceptable parameters
3. a most excellent way to test how far fully autonomous development can go -- in particular, to test how the "rest of" one's configuration/scaffolding/setup is, for such "auto builds"
The whole idea with auto-YOLO and Ralph seems to be you loop a lot and see what you can get. Very low effort, surprisingly good results. Just minor variations on branding and implementation.
Either way, in my experience, auto-YOLO can actually work pretty well. 2025 proved to be cool in that regard.