Qoder: My AI Partner Designed, Coded, and Tested This App for Me.

This entire video was generated from a single sentence of test. The AI to do this is out there, but it's often locked behind walls of complex code.

To build an app for this, you'd normally spend hours just reading dots like this.

So today we're going to build a simple user-friendly video generator from scratch. I will use an AI coding tools

scratch. I will use an AI coding tools called coder to see if we can streamline this whole process. First, let's take a quick look at my MCP setup. As you can

see here in my settings, I've enabled a few like context 7 and sequential syncing MCP servers, uh, which gives the AI it advanced reasoning abilities.

Okay, with that setup, the first step in any project is giving the AI context.

Now, in this case, from best practice, I found that the fetch MCP cannot directly assess the API uh API file because of

login restrictions. So, I've done a

login restrictions. So, I've done a quick manual preparation.

Um here I've summarized the key information from the API dot into this simple markdown file with the help of Google Gemini. So I'm now adding it to

Google Gemini. So I'm now adding it to the rules section of coder. Think of

rules as permanent instructions. The AI

will treat this file as source of truth for our entire project. Now that coder understand our API, we can start building. Let's kick things off with

building. Let's kick things off with coder's main feature from building from scratch. Quest mode. I'm going to give a

scratch. Quest mode. I'm going to give a highle mission for the entire applications.

So build a complete radio web app that provides a user user

friendly interface for the violin text to video API based on the instructions

in the rules file.

And um I'm telling him my API key is stored in uh env file.

And uh here's the first major difference I've noticed compared to a workflow in a tool like cursor. Um instead of just

creating files and then asking for code, coder start with a technical design phase.

This is really interesting. Before

writing any code, um, it generated a complete technical specification based on the rules I gave it. Uh, this

planning setup is great for complex task to make sure the AI and I are aligned on the architecture. This plan looks solid.

the architecture. This plan looks solid.

So, let's tell it to proceed.

We might press start. Now,

[Music] now we are in the action flow phase.

Coder is executing the plan it just created. And the to-do list here

created. And the to-do list here um shows its step-by-step process, which is a nice way to um track progress.

So here um the AI is autonomously working through its checklist. It's

generating the project structure, writing the backend service, uh building gradial UI and even generating unit test for um the service it created. Now that

all the steps are done, um the quest automatically move to the final phase, the task report.

Um this is another a very interesting feature. KOD provides a structural

feature. KOD provides a structural summary of the entire task, an overview of the changes, test results, and um

diff of every file that are modified. So

it's a formal code review setup for the AI work. So

AI work. So uh at the end I will accept the changes and try to run the

application here.

I'm trying to make a uh final run here with all the uh command line setup.

So I just guess so let's try to uh paste the designed prompt here and to try if we

and generate a beautiful video just like the beginning shows.

Yeah. So, back to the terminal. uh it

seems that uh uh response status is uh indicate that it's in the right way and the script is pulling response from

from bian API.

There you go. So um this is our final result. Uh we can see we from a single

result. Uh we can see we from a single high level prominent one rule file we have a working AI video applications and

the process was very structured design execution and been well reviewed. So our

app is built and working but the project isn't truly done until it's documented.

So what if a new developer joins the team? So this is where a feature encoder

team? So this is where a feature encoder called rile wiki comes incredibly useful useful

here. As you can see here ripple wiki

here. As you can see here ripple wiki can analyze an existing codebase to a document. So this is easy for someone

document. So this is easy for someone who is new to this project to get familiar with the code in a very short

time. So to sum up for building a new

time. So to sum up for building a new project from a single idea coder quest mode is uh introduced a very

structured plan first approach and once the work is done ripple wiki can help document the result. It's a compiling

end to end workflow. So this is my first look at coder. Uh let me know in comment what you think of this structured approach to AI development

and I will put uh all the links and the GitHub repo of this project into the description. Thanks for watching.

description. Thanks for watching.