Оффлайн
- Регистрация
- 1 Мар 2015
- Сообщения
- 1,481
- Баллы
- 155
AI is here, it’s in your phone, your laptop, your devices and even your toilet, for that matter, but more importantly, it’s in your code editor, and as a developer, that’s the only thing that is going to matter
Vibe coding is good and all, but you need to understand how to use AI to your advantage and not to believe all the stupid things it comes up with
So here are things you should do if you are not when you are coding with AI by your side
1. Know your source code
Around 41% of code on Github is AI-generated, and I am sure most of that will be just copy-pasted, generated and never even looked at once by the people who wrote it.
So, my first advice is to know exactly what your source code does. Many people don’t even know what the AI is telling them, even though they know it can hallucinate.
So, don’t blindly trust the code the AI gives; know what the source code is about and understand how it works.
If you are just asking Cursor to build a SaaS app from scratch with just your idea, it’s of course going to choke; understanding the basics is as important as the idea
2. Draw it out
Diagram it out. Get a piece of paper and pen and just draw out how exactly the data in the application is going to flow and how the user is going to interact with the website or the project, or if it is a backend application, know-how the data it is going in the database and just work it out.
This works because diagrams reduce your cognitive load. Your brain can’t multitask, no matter how much coffee you drink, so diagramming it out just lowers the amount of stuff your brain has to juggle. It doesn’t have to store the idea you have, or what to have for dinner, or why your ex liked that one Reel but didn’t respond to your text. Even napkin doodles help.
3. Debug Your Brain First
Curse of False Clarity plagues all; everyone feels they have understood a particular thing, but the matter of fact is, you know nothing; there is this false sense of hope that your brain wants to latch on to.
In fact, this is the same feeling you get when you watch a tutorial, even if you followed it along, it won’t matter unless you make a project yourself, solve a problem yourself, or even talk about it to others, you’ll know nothing.
Try explaining out loud to your pet or a rubber duck if you know what I mean. If your imaginary cat looks confused, that’s a sign
4. Use AI, Don’t Worship It
Treat AI like a caffeinated intern: helpful, fast, and more often than not wrong. Use it to brainstorm ideas, unblock yourself, rephrase confusing code, or even add comments. But if it writes your entire project, you didn’t solve a problem.
, let it write code. Let it generate helpers, test cases, and boilerplate. But never let it think for you. Ask it questions. Push back. Challenge it. The more intentional you are, the more AI becomes a turbocharger, not a crutch
5. Reverse Engineer the Vibe
If you have already vibe-coded something that “works”, don’t stop. Now it’s time to tear up the code and perform a post-mortem. Go line-by-line and ask the following questions:
It’s time to become a manager of your code. Break it. Rebuild it. This turns AI gibberish into actual skills
6. Vibe Fast, Refactor Slow
Use AI to move fast. That’s the whole point. But don’t ship raw vibe. Your job is to take the chaos it created and sand it down into maintainable code.
Make an actual checklist:
Remember this code is going to be on the internet long after and is going to be used for training the model on further code, so the bad the code, the bad the training.
7. Make AI Teach You
Before you accept any AI code, ask it:
Make it justify its answers. Think of it as Socratic coding. You are not a passive student - you’re a slightly annoyed teacher
8. Don’t Skip Setup — Scaffold With Intention
, AI can vomit the project structure in 0.4 seconds. But that doesn’t mean you should accept its idea of architecture
Spend a few minutes planning:
If you start with chaos, you’ll end with chaos. Vibe responsibly
9. Reinvent With Limits
AI makes it feel like you can build everything at once—An entire SaaS platform before breakfast.
So add constraints:
That’s when your vibe hits product velocity
10. Reuse Your Vibe Code Bank
Keep your “vibe archive” of every decent snippet AI writes. Save it, categorise it, and make it part of your stack.
Stop regenerating the same auth boilerplate 15 times for 4 different projects.
Vibe coding is good and all, but you need to understand how to use AI to your advantage and not to believe all the stupid things it comes up with
So here are things you should do if you are not when you are coding with AI by your side
1. Know your source code
Around 41% of code on Github is AI-generated, and I am sure most of that will be just copy-pasted, generated and never even looked at once by the people who wrote it.
So, my first advice is to know exactly what your source code does. Many people don’t even know what the AI is telling them, even though they know it can hallucinate.
So, don’t blindly trust the code the AI gives; know what the source code is about and understand how it works.
If you are just asking Cursor to build a SaaS app from scratch with just your idea, it’s of course going to choke; understanding the basics is as important as the idea
2. Draw it out
Diagram it out. Get a piece of paper and pen and just draw out how exactly the data in the application is going to flow and how the user is going to interact with the website or the project, or if it is a backend application, know-how the data it is going in the database and just work it out.
This works because diagrams reduce your cognitive load. Your brain can’t multitask, no matter how much coffee you drink, so diagramming it out just lowers the amount of stuff your brain has to juggle. It doesn’t have to store the idea you have, or what to have for dinner, or why your ex liked that one Reel but didn’t respond to your text. Even napkin doodles help.
3. Debug Your Brain First
Curse of False Clarity plagues all; everyone feels they have understood a particular thing, but the matter of fact is, you know nothing; there is this false sense of hope that your brain wants to latch on to.
In fact, this is the same feeling you get when you watch a tutorial, even if you followed it along, it won’t matter unless you make a project yourself, solve a problem yourself, or even talk about it to others, you’ll know nothing.
Try explaining out loud to your pet or a rubber duck if you know what I mean. If your imaginary cat looks confused, that’s a sign
4. Use AI, Don’t Worship It
Treat AI like a caffeinated intern: helpful, fast, and more often than not wrong. Use it to brainstorm ideas, unblock yourself, rephrase confusing code, or even add comments. But if it writes your entire project, you didn’t solve a problem.
, let it write code. Let it generate helpers, test cases, and boilerplate. But never let it think for you. Ask it questions. Push back. Challenge it. The more intentional you are, the more AI becomes a turbocharger, not a crutch5. Reverse Engineer the Vibe
If you have already vibe-coded something that “works”, don’t stop. Now it’s time to tear up the code and perform a post-mortem. Go line-by-line and ask the following questions:
What does this do?
Could I write this myself?
Is this needed, or just AI filler?
It’s time to become a manager of your code. Break it. Rebuild it. This turns AI gibberish into actual skills
6. Vibe Fast, Refactor Slow
Use AI to move fast. That’s the whole point. But don’t ship raw vibe. Your job is to take the chaos it created and sand it down into maintainable code.
Make an actual checklist:
Is it readable?
Are the variables named like something humans would use?
Do the comments make sense?
Remember this code is going to be on the internet long after and is going to be used for training the model on further code, so the bad the code, the bad the training.
7. Make AI Teach You
Before you accept any AI code, ask it:
“Explain this like I am five”
“What are the edge cases?”
“What are the potential bugs here?”
Make it justify its answers. Think of it as Socratic coding. You are not a passive student - you’re a slightly annoyed teacher
8. Don’t Skip Setup — Scaffold With Intention
, AI can vomit the project structure in 0.4 seconds. But that doesn’t mean you should accept its idea of architectureSpend a few minutes planning:
What goes where?
How will files talk to each other?
Is this over-engineered, or can it be maintained after two days?
If you start with chaos, you’ll end with chaos. Vibe responsibly
9. Reinvent With Limits
AI makes it feel like you can build everything at once—An entire SaaS platform before breakfast.
So add constraints:
Set timers
Use MVP thinking
Force yourself to launch something bad
That’s when your vibe hits product velocity
10. Reuse Your Vibe Code Bank
Keep your “vibe archive” of every decent snippet AI writes. Save it, categorise it, and make it part of your stack.
Stop regenerating the same auth boilerplate 15 times for 4 different projects.