Disclaimer

This guide is fully hand-written, there is zero prompting involved in the thinking or creation of this guide, i have the belief that AI could never express my own thoughts and lifestyle as good as me no matter how well he is prompted

Everything shared in this guide is simply my own experience and views, please think for yourself and see if this guide speaks to you, if it does i would love if you shared it to make more people aware of it

Why should YOU read this guide ?

To keep it short i have been using AI 8h+ per working day (without exaggerating) for a lot of months now, before i started trying it i've been somewhat against full agentic workflow like a lot of coders who spent years in the industry, it just felt like getting robbed of my hobby and that it just became uninteresting

From where i stand now, i would almost never go back to normal coding, the entire point is that an agentic workflow just allows you to 10x if not 20x your "work power", you will get to achieve literally ANY project you want under 6 months, where before it could have taken more than a year or two to code extremely complex softwares

Following this guide will allow you to (almost) fully rely on the AI to code for you, you will become a "leader/manager" rather than the programmer, you need to ditch the idea that you will code anything and rather embrace your role to dictate the AI exactly what to do

If you wish to reach 100% usage on all your subscriptions every month and get true work done with AI this guide is for you

You will learn about multiple "AI prompt hacks" as well as physiology-based ideas i use daily to improve my workflow to be able to withstand an entire full day of non-stop working, up to 16+ hours without "pharmaceutical aids"

I have not written a line of code manually ever since i started working this way
You envision the code and the way you would code it, the AI codes it for you through a prompt, this is the vision you must have

Of course i don't claim to have achieved the best way to prompt AI or use it, it is also very model dependant, for reference i have been using Cursor so i use a mix of Claude/ChatGPT/Grok (previously even Composer)

I am making this guide from my every day usage and most likely people who actually worked in improving AI's have a better understanding on how to prompt than i do
If you are one of them feel free to get in touch with me on X, i'd be more than happy to discuss / improve this guide for others

Enough talking, let's get to it !

Is there any limit to agentic coding ?

There is almost no limit to it, you can do things well beyond your current understanding, remember you are a "leader/manager", generally programmers own the technical skills while the leader just requests without fully understanding what it means in terms of implementations

This is similar except even better because YOU might actually have the skillset and understanding that your manager did not and even if you do not it will usually never be worse than what companies used to be in terms of idea to output

You become more of a "translator", the idea is there and you formulate it in the best way possible for the AI to understand and implement

This is not a reason to stop learning, we will touch on that later in the guide

Who can learn from this guide ?

Complete beginners as well as extremely advanced coders / AI users, you will always reshape your understanding of AI by reading through it !

The steps are in order to setup a proper agentic workflow in your directories and how to live with AI, i will later share how i manage my time and how you can "never work" again

Optimizing your AI workflow in 3 steps

Step 1 (Setup)

1. Download your favorite IDE with AI integration, it can be Cursor, VS Code, Claude Code, ChatGPT, ...
2. Think of the AI as a partner, someone who might be more skilled than you overall but may be less skilled in specific areas, which is why you must understand that proper prompting is EXTREMELY IMPORTANT, put yourself in the shoes of the AI on every single prompt and try to view how easy it would be to understand what your "boss" expects from you, if it isn't then refine the prompt further and give more context
3. Proper setup of the environment, this is so underrated and the most important step and a LOT of people do not bother with that, i can tell you this makes your experience 20x better, let's get into practice

Practice (Rules, Skills and MCPs)

Whenever you make a new project (or integrate AI in an existing one) it is extremely important that you think of all possible "ends" of the project and how the AI could drift off from what you want, this is exactly what you want to gate

First off start with basic things such as your coding style, this is where the first good practice comes in you must almost ALWAYS make your idea go through the lens of the AI, as i mentioned before he most likely knows more than you do in general areas (how to make good rules is the example here)

• Prompt example: I need you to build a code style rule from (attach your folder path/url) and apply it as a rule for this repo
  Not the best example, continue reading

This goes for ANY rule you desire to do, rules specifically are very important to put through the AI lens
Ensure you review EVERY prompt you are making because you can usually find a better way to formulate it the second time:

• Prompt example: I need you to analyze the code style fom (attach your folder path/url) in depth and apply it as a rule for this repo

This is a much better type of prompting, no redundancy and precising that we need depth
Of course you also need to think if this rule stands as an user rule or as a repo rule, usually code style and so on better fits into repo's themselves if you are working on multiple projects with different code styles

Here are example of prompts i have done in the past for rules setup:

• Prompt example: make a new rule for tests, we need to ensure the tests that are temporary get deleted afterwards or cleaned whenever possible inside "tests\temp" the long term tests need to sit in one directory "tests" and must be kept clean as well, keep the tests minimal if possible it must target exactly what we desire

• Prompt example:
  we need to have coding style rules so that it does not go out of hand quickly, you can copy the coding style from: (repo_path)
  
  just make sure we are using helpers as little as possible, we usually want code to be cleanly implemented inside their own functions, if a helper is needed ensure it is fully able to be shared from other components, and store it inside "helpers" folder or something like this
  copy the folder structure as well from (repo_path)
  
  you can add good practice (language you work in, f.e C++) rules from (whatever company you like the rules of)
  its very important that the repo is kept clean, after each major change see if we can rework the code into tighter parts / less redundancy
  
  its VERY important that you remember that simplicity is key and we must strive for it, we must avoid overcomplicated systems if they can be minimized while keeping the exact same outcome

Be extensive with your prompts, ensure they are long enough and have the required context, this is how you avoid "slop" (we will talk about it right after)

For the skills you can apply the same idea, except you need better depth in your prompt:

• Prompt example: I need you to create a skill that agents will use to work on MySQL, it is important to go in-depth with the MySQL possibilities so that the agent using this skill will always find the best way to implement something

I am not very familiar with MySQL this is why i am being broad, remember the AI likely knows more than you do unless you are very skilled in a specific field, you may also beforehand ask what the pitfalls are or what are behavior we would not want to encounter while coding using MySQL for longevity and stability of the implementation and then you can shape the prompt for the skill around the answer the AI gives you

I encourage the use of MCPs that actually help the environment, it is an extremely good way to add more details to your prompting / workflow and let the AI use tools that can be custom made, the main way i am using MCPs is either to help within the AI environment or let the AI interact with certain softwares on the computer

There are so many uses to MCPs and they are extremely underrated, just make sure you do not download slop MCPs that end up being useless or burn your tokens

Step 2 (Avoiding slop in a world full of slop)

You must have already heard or seen "vibecoded" or "ai slop", this is almost everywhere in a world where anyone can pretend to be good at something but not everyone is good at using AI correctly, far from that

It is extremely important that you become careful with the rules / skills / MCP you implement from the web because A LOT of these can make your AI usage MUCH WORSE !!!

Usually you are better off making your own rules / skills / MCP tied to what you noticed your workflow lacks, it takes more time but evolves your workflow so much more in the way you exactly want it to, this is something i would 100% recommend to anybody working with AI
For that reason i will not share my own rules or skills publicly because they are tied to what i am doing and will most likely not fit for your usage

On that note, i will share an MCP that i made (coded in Zig to be compilable on every OS, coded strictly with a full agentic workflow, i did not write one line in this repo)
I believe this MCP improves AI a lot over the longevity of a project, it is called project progress

The entire point of this MCP is that the AI will update the prog