Audience¶
You'll find this product useful if you want to build software and you are in charge of defining how it should behave.
Problem statement¶
You'll be especially interested in this product if any of the following is also true:
- you've tried AI coding agents and they sometimes
- don't do what you meant, or
- start doing stupid shit after chatting for a while, or
- refuse to help you, or
- don't test the software properly;
- your intended software is mainly non-visual (e.g., scrapers, bots, complex backends, dev tools, and basically anything but a frontend);
- you either don't know what context management is, or you are tired of it.
All those situations come from two base premises:
- Putting your intent into unambiguous text is harder than it seems.
- After you've successfully done that (if you ever do), turning that into working, polished, and shipped software usually isn't as seamless as just clicking a button.
The proper solution¶
Just Ralph It, or JRI for short, is the solution to those problems packaged as intuitive-to-use software, that doesn't require you much more other than simply sharing your thoughts through a single chat, without having to think in terms of context, sessions, orchestration, backpressure, and ideally any AI engineering term you can think of.
The name comes from the Ralph technique, by Geoffrey Huntley, though now that is expected to be just one component of a larger system. Worth mentioning that Geoffrey himself pointed out that productizing the Ralph technique would not work. He argues that it's highly supervised, as for Ralph to work you need to babysit it; that you need to have an engineer on the loop, tuning it like a guitar. Even Anthropic has already tried it with a plugin and with the popular /goal, none of which solved the problems mentioned above.
Nevertheless, JRI is my attempt to do it better, to craft the guitar that comes tuned out-of-the-box. My thesis is that LLMs are already—and if not, they will definitely be—able to make this possible, if given the proper system around them, one which handles the inherent limitations of LLMs, such as infra, context size. Inherent is an important word here, because it means that no matter how smart an LLM may be, they can't solve those limitations solely by themselves. No doubt, LLMs' smartness makes it easier to address them, but we still need JRI or a JRI-like system.
How it works¶
This is a public work in progress. For the latest overview, send this to your favorite AI agent, or just inspect the repo yourself:
Study https://nicolaspujia.com/just-ralph-it.md and its GitHub repo. Then, in plain English, explain where it stands today and how what has been implemented so far works.
Success scenario¶
JRI works as expected if I can define JRI itself with little friction out of my mind, click "Just Ralph It," and see a working clone emerge that behaves as expected.