₦airaland Forum

Seun:
Creating software engineering tools for personal use is one of the most wonderful uses for vibe coding. The AI models are like us. They make a lot of little mistakes, which is unacceptable for tasks like code refactoring, but they can solve those tasks using precise & accurate scripts.

I am Already trying to learn this

If you're building AI voice agents that need to operate in Nigeria, ConnectSIP offers SIP-ready virtual numbers designed specifically for modern AI infrastructure. Worth checking out if you've been hitting the same walls everyone else encounters with traditional providers.

Nice one ☝️
I vibe code also

Interesting

AI is the best hing over the years.

If your coding agent keeps trying to write code when your plan is not totally ready, use the three ways trick. "Show me 3 ways to tackle this problem" If you just ask for the solution it will pick a half-baked solution and immediately try to implement it without your permission.

Seun:
If your coding agent keeps trying to write code when your plan is not totally ready, use the three ways trick. "Show me 3 ways to tackle this problem" If you just ask for the solution it will pick a half-baked solution and immediately try to implement it without your permission.

Thanks

LLMs are great at translating from one human or programming language to another. I ported my AI Agent written in Python to Go, Rust, Typescript, and Java over a couple of days. Now, whenever I change the Python version, the AI can translate the change to the other languages. My intuition tells me that maintaining a codebase in two languages (not 5) can lead to better code. I'm thinking, Python plus Rust.

Can you explain to me why my comment was banned despite contributing to the subject matter?

Seun:
LLMs are great at translating from one human or programming language to another. I ported my AI Agent written in Python to Go, Rust, Typescript, and Java over a couple of days. Now, whenever I change the Python version, the AI can translate the change to the other languages. My intuition tells me that maintaining a codebase in two languages (not 5) can lead to better code. I'm thinking, Python plus Rust.

wazzihu:
Trying to learn AI automation, can you recommend a good source

Use N8N

idalex:
@Seun m in support, t’s high time we get AI section over here on nairaland

@ Zaksart kindly reply m pm

Sorry for the late response... I just did.

Anyone excited by Claude 4.5? It's supposed to be better than even Claude Opus 4.1 which costs 5 times as much (a ridiculous price)

Matoss:
Use N8N

Interesting. What have you automated with n8n?

Following

Updated AI tool rankings

Cursor: Best IDE. Best autocomplete, excellent model selection, best UI for diffing and change management, new agent features every day.

Claude Code: Best command line agent. Usable not only for coding but for server management. Poor frontier model selection (only Claude).

Codex: Most improved. Best UI selection - VS Code plugin, cli, and cloud agent. Decent model selection. 5, 5-codex, 5-mini, high/med/low.

Seun:
Interesting. What have you automated with n8n?

For me, I just started learning it.

On Sunday I created an account with Notion as a cookie company with a customer base in various locations across the UK. Then set up a trigger on n8n to reply customer bookings.

Meaning if a customer places an order my n8n would confirm the order send a message to the customer, confirm payment and send me a notification confirming the payment. Saving me hours of having to do it manually.

Below is a Pic of the company profile I created on Notion. It's not real though. Just something I'm practicing with.

N8n to my understanding is primarily designed for users who want to automate tasks and integrate various applications without extensive coding knowledge.

Just have an understanding of what you want to automate and build your workflow towards it and the app would execute it for you. Saving you precious time.

While Prompt engineering is cool, I just figured out that reverse engineering works better and gives better result than any prompt you can write. I have used it to build about 5 different apps for clients. Its very cool.

lovelybobo:
While Prompt engineering is cool, I just figured out that reverse engineering works better and gives better result than any prompt you can write. I have used it to build about 5 different apps for clients. Its very cool.

That's for UI, right? You give the AI agent a picture of a similar design or a frontend and then it recreates it for you? And then you tweak the design?

The last time I did this, gpt5 thinking on ChatGPT did the best job. Claude Opus 4.1 was surprisingly bad. Pls share your more of your ideas.

Seun:
That's for UI, right? You give the AI agent a picture of a similar design or a frontend and then it recreates it for you? And then you tweak the design?

The last time I did this, gpt5 thinking on ChatGPT did the best job. Claude Opus 4.1 was surprisingly bad. Pls share your more of your ideas.

Let me share an example from one of my Apps.
The concept is very simple, The idea is to make AI write its prompt first and thereafter, it will execute the prompt.

This is from one of my projects

Take the role of prompt engineer with 20 years experience. I will you give my desired results
and you create a detailed prompt for generating these results.
The name of the app is Physical game monitor App
.
It will have the following features


1 input button for name of users and the inputted names appear on a grid below,
2 The grid for names will have three names in a single row
3 For example 12 names will appear in four rows
4 Use this code as an example and also a guide for the grid " "
5 Each item in the grid will have four control buttons - E for Eliminated, W for Won, S for Suspended, C for Cautioned
6 When a person is eliminated, The item containing the name will zoom out and a red X strike appears on it as it returns to it original position
7 When a person has won, The item containing the name will zoom out and fireworks begins to fall until the name returns to it original position
8 When a person is Suspended, The item containing the name will zoom out and the color of the name will change to red
9 When a person is Cautioned, The item containing the name will zoom out and the color of the name will change to Amber
10 All names will appear in green from the beginning
11 Add any other feature that will make the app very interesting and easy to use.

I added other features much later

I just tried genspark ai

arejibadz:
I just tried genspark ai

Is that a ChatGPT clone?

Update: Genspark is an agent also, just like Manus

AI Vibe coding helped me, with my medium amount of knowledge on CSS, JavaScript and HTML, I was able to build a self-help web app assisted by Claude, it was a really really nice but tiring experience.

Kimcutie:
AI Vibe coding helped me, with my medium amount of knowledge on CSS, JavaScript and HTML, I was able to build a self-help web app assisted by Claude, it was a really really nice but tiring experience.

It's so tiring when you are not familiar with the platform it's coding for you. My initial experience in Android was bad. The AI kept making mistakes that I didn't even understand. But in the process I learned the basics of Android (gradle, libraries, manifest, permissions, etc)
and I feel more confident to vibe code for it. When I go back to android I'll probably have to feed the AI with up-to-date documentation.

That's meta prompting. I use it too. It's a game-changer. It's value is in saving you from having to write what the AI already knows. If I want to code a todo list, I only have to specify what makes my version of todo list different from what the AI thinks a todo list should look like.

If not for this technique, we'd have to wait for the AI to write the code and test it before asking for changes, which is very expensive and slow.

My technique is a little bit simpler than yours. I just say, "write a (detailed) prompt for an ai to: _____" or "prompt an ai to: ____". Then I read the generated prompt and if there's a part I don't like, I'll modify the meta-prompt to make it generate that aspect the way that I want it. Instead of progressing the conversation I keep going back and adding new specs to the meta-prompt until I'm satisfied with the generated prompt. My meta-prompt ends up looking like yours, with the list of features and clarifications, after a few iterations of regenerating the prompt.

lovelybobo:
Let me share an example from one of my Apps.
The concept is very simple, The idea is to make AI write its prompt first and thereafter, it will execute the prompt.

This is from one of my projects

Take the role of prompt engineer with 20 years experience. I will you give my desired results
and you create a detailed prompt for generating these results.
The name of the app is Physical game monitor App

It will have the following features
1 input button for name of users and the inputted names appear on a grid below,
2 The grid for names will have three names in a single row
3 For example 12 names will appear in four rows
4 Use this code as an example and also a guide for the grid " "
5 Each item in the grid will have four control buttons - E for Eliminated, W for Won, S for Suspended, C for Cautioned
6 When a person is eliminated, The item containing the name will zoom out and a red X strike appears on it as it returns to it original position
7 When a person has won, The item containing the name will zoom out and fireworks begins to fall until the name returns to it original position
8 When a person is Suspended, The item containing the name will zoom out and the color of the name will change to red
9 When a person is Cautioned, The item containing the name will zoom out and the color of the name will change to Amber
10 All names will appear in green from the beginning
11 Add any other feature that will make the app very interesting and easy to use.

I added other features much later

Seun:
That's for UI, right? You give the AI agent a picture of a similar design or a frontend and then it recreates it for you? And then you tweak the design?

The last time I did this, gpt5 thinking on ChatGPT did the best job. Claude Opus 4.1 was surprisingly bad. Pls share your more of your ideas.

Gemini is excellent at this, you can give it a try