Looking for programmers is difficult and painful. Even with a good budget.

Usually answer briefly - not interested or just will not take, and freelance exchanges are filled with writers of scripts for 200 lines of code, if you know what I mean.

But since we live in a wonderful time of neural networks, I spent a couple of evenings looking for an executor, and then I spit and sat down to write it myself. As always I can't get my hands on a full-fledged video, but maybe my experience of a year of neuroprogramming in text form will come in handy to write almost anything myself. No kidding, the screenshot is a Telegram combinator for thousands of lines of code that juggles half a dozen apis to get the result I need.

However, to write it, I had to go through a thorny path, where I experienced what tech debt is, and I had only heard about it before. I started writing with a single file, the file grew, the answer of 300 lines of code (exactly that much the AI can write in a single message) quickly became insufficient, the expansion of functionality piled up huge files, which at some point became impossible to update and maintain.

To avoid repeating my mistakes you can follow the following rules:

• No ChatGPT, paid or free it does crap, the best and only one Claude.ai
• Immediately create a project (GPT has a limit of 20 files, I haven't met any limitations here yet). A project is a set of files from future development. In the settings we specify how to give a response: briefly, without comments in the code to save tokens, one file in one response, if we plan to fix several files - specify which ones first.
• Depending on the programming language and the type of project, I highly recommend creating a chat room and asking about the structure of the future project. If I had done so, I would have learned at once that telegram bots have handlers, keyboard buttons are stored in a separate file, and each module should be written in a separate file.
• The main rule is to never write a large functionality at once, one chat +1 simple function. Even now, I can understand where it is worth giving a task to several functions, but at first it is extremely difficult to track what and where breaks or works wrong.
• To work with the API, you just need to show an example request and an example response (most often json), and they are almost always in the documentation.
• To get a good result, you need to clearly understand what steps there are in the process and what follows what. You don't know? Give input and ask the AI how it would solve the problem.

A couple of weeks of practice and you will write automation of any complexity, and in any language, no course will teach you that 🙂 🙂