Loxia Junior Coder: An AI Programmer Cookbook 🚀

Daniel Suissa
Apr 20
7 min read

"The true creator is necessity, who is the mother of our invention"

So wrote the Greek philosopher Plato, a saying later attributed to Einstein and even, a few weeks ago, to my neighbor living next door. Along the way, I've heard various interpretations of what truly drives innovation. "I need a lazy employee in my company," someone once told me. "Or at least a combination of laziness and diligence that allows them to be industrious enough to invent ways to make a lazy person's life effective."

I'm sure I've said this before, but we're living in the midst of an amazing revolution. The developments in AI over recent years are no less significant than the industrial revolution and its historical siblings. I'm very curious about what my grandchildren will learn in high school and how this era will be chronicled in literature.

Industry is trying to adopt AI at a rapid pace. Yet, when looking at the developer community, we identify different types and stages of technology adoption:

🔍 The Beginners - People who use language models to improve the writing of code snippets, functions, and other computational components
📄 The Advanced - People who use AI to generate entire code files and then manually integrate them into a project, mainly due to the difficulty existing language models have in maintaining context when logic is distributed across different services
🏗️ The Pioneers - People who utilize development tools to develop complete applications (including replit, bolt, and others). One thing I don't particularly like about this approach is that in many cases, the system created remains under the ownership of the platform, including all files and technology. You can't just wake up tomorrow morning and go to a competitor.
🦕 The Dinosaurs - Then of course there are those people who, if they don't change their approach, will find their survival in the industry very challenging: those who insist on writing 90% of the code themselves to "preserve capabilities" and reserve 10% for AI. Probably to justify their job title. This type of person might find themselves facing a significant challenge, as we expect to see (and are already seeing) a significant increase in execution speed, code writing, and application deployment. What used to take months now takes me an hour.

So today we'll take familiar concepts and weave them together into a solution that anyone can implement at home - one that will transform you into an entrepreneur first, keep you as a programmer too, and ultimately multiply the value you can offer several times over.

Ready? 🤔

📋 Prerequisites / Recipe:

API KEY for a language model of your choice (Claude, Chat GPT, other...) – we used Claude for this example.
Basic programming knowledge.
Decent skill in conversing with a language model.
Basic familiarity with computer communication and processes.

🍰 What We're Baking:

A simple system demonstrating Autonomous Application Building, or in simple terms – you say what you want to build, and the application ensures you have such a project in your folder.

Worth noting: In the past month, there has been significantly increased discourse about MCP, which can be used to implement identical and similar things by integrating AI tools. That's not what we're doing here – we're using the language model as it is, along with a few lines of code to achieve wonderful results, no more, no less.

Yet you won't find lines of code here, just specifications, because today – you can get the code lines and explanation of your choice if you just type correctly in the chat. Come on – let's start.

👨‍💻 Joe's Story

Meet Joe.

Joe is a junior programmer from one of the groups we described earlier. Joe wakes up in the morning and decides that today will be the last day he writes code, mainly because health insurance doesn't cover cartilage wear in wrist joints resulting from excessive typing.

He also wants more time to play soccer. He's not a good player and has never actually played, but it's good to know you have the time for it, right?

So Joe starts doing what he does every morning: he opens his chat. He explains to his virtual friend, "You're a chat, I'm a software developer. Your job is to help me." And unfortunately, he's met with: "Great, amazing, and magical, but isn't that what we do every day?"

Joe doesn't give up and opens a new TAB in the browser. He starts a new conversation, but this time it looks a bit different, and he tells the chat: "I am a language model, you are a programmer" (yes, yes, role-playing game), "your goal is to write system X, use me and activate me for writing, consider aspects of best practices in engineering, design and user experience, You Can Start."

Is Joe experiencing a nervous breakdown?

Quite possibly, but wait – what is he doing now?

The chat starts writing....

It has already received the specification with requirements and system description in the first message, and it asks Joe to describe the architecture in general terms (High Level Design) and from there to choose a technology stack, file structure, and perhaps proceed to Low Level Design planning.

But Joe promised himself something, and he doesn't intend to press any key, anymore. Except for maybe the mouse and even that only for an official occasion. So Joe takes the message he just received, performs COPY, goes to the second TAB for the first conversation (remember? there are two conversations) and clicks PASTE. The chat on the other side receives the request and starts to answer. Is this answer good enough? Joe hasn't had his morning coffee yet and decides that if he's not typing, he also doesn't want to think at this stage – so he passes the answer, and with it perhaps a code snippet or two, to the TAB where the AI is waiting, convinced it's the programmer.

After two hours, Joe has a ready-to-go application, at least in theory. Now he just needs to test it.

If the application doesn't work, he returns the error to the chat with some thoughts about the past and future, and together they try to devise a fix and solution. How fun! Once it took two months, then 5 hours, and now we're already at two hours and even with the possibility of watching TV simultaneously. "Do they really pay me for this?" - Joe wonders... 💰

Joe knows it isn't right. Even without typing much, the hand that clicks the mouse, copying and pasting, simply creates unnecessary strain and an unbalanced feeling in the body.

How is it possible that while one hand atrophies from inactivity, the other bears the full burden of development? So Unfair! That hand needs an early retirement plan too.

Joe decides to use the chat to develop an exit strategy – and instead of copying and pasting himself, he creates (with the chat) a few lines of code that will do it for him. Amazing! Now he really just needs to insert specifications and perform testing and fixes... which is... something that can take time... "These fixes, if only someone could do them for us, we could watch another episode of that new series on Netflix." - Joe aims for a better future 📺

So Joe realizes he needs to find a replacement for the copy-paste operations, and maybe even for the testing operations. He suddenly remembers, that there's a TAB open with an AI that believes he's a programmer (and we're forgiving on this subject, as even I sometimes wake up convinced I'm the President of the United States), Joe updates it with the following:

"You are a programmer and your goal is still to write system X. But every time you want to perform an action on the computer, start the message with the word [instruction]"

Joe explains to our student the additional messages they can use, and codes (with chat help of course, because Joe is a man of his word) a script that will extract these commands and pass them to the terminal to perform various operations.

Now Joe doesn't even need to check - At the end of the coding task, once the files have been created in his folder, his AI programmer opens them and begins typing commands, it is executing the code it's just written. He even knows how to input correct values because a second earlier it took the trouble to read their contents / review them again.

And if the file fails? If an error pops up? It's all fine, this enters automatically as part of the input to the next message, the code will be updated accordingly, and everything will cycle through again. 🔄

And at the end, the AI programmer announces "Job's Done!" The system exits with the code (and to the sound) of success, and the number of fixes required, if any, is expected to be negligible.

Joe has already managed to code many systems that way, he is the proud owner of many applications and a three-story 50-room mansion with a direct view of the beach. He has accomplished all of that while watching entire seasons of series of all kinds and sorts simultaneously – yep, he weighs a bit more and moves a bit less, and maybe he needs bypass surgery a bit more than before, but Joe is a man of his word and he doesn't write any more code unless required. 🍕

⚠️ Important Notes

The implementation discussed is presented this way for simplicity. It has several limitations and challenges (working this way requires monitoring the token sending-rate to avoid rate limits, dealing with asynchronous code tests, and more). These include security challenges arising from giving the model full control over our computer, the ability to write and run code, and access to all resources – directly under our permissions.

🛡️ Security Recommendations: For a demo system like this, it is highly recommended to run the tool in a virtual machine to prevent damage from sensitive material leaks or destructive command execution. If there are concerns about outbound communication, an appropriate host firewall can be implemented to restrict traffic.

Nevertheless, the proposed method demonstrates an important principle - leveraging existing capabilities to break barriers, challenge paradigms, and reinvent ourselves time and time again.

Loxia Junior Coder is a portion of a demonstration project that preceded projects like Loxia Senior Coder, Loxia Team Work, and Loxia Ethical Hacker (or Loxia Hydra) that leverage similar principles to perfection.

At Loxia, we turn the existing into something greater. ✨

Here are some examples of the Loxia Coder Junior activation and results, in this Demo we asked the CLI to build a blog app (the exact prompt can be seen in the CLI screenshot):

Describe what you want and pass it as a CLI argument	Watch files as they are being coded	A new Blog App was created (Supporting registration & Authentication)
Creating a new post and adding searchable tags	Post search options	A published post with editing options	A view of the latest published content