Sleep? Never Heard of It: How I Built an Entire App with AI in One Weekend
SubtitlesKing.com is how you add subtitles to any video for free
I recently spent a whirlwind 48 hours building a brand-new application with the help of AI. In non-coding terms, in two days I was able to lay a cement foundation, build a wooden house, insulate it, run the wiring, put in the plumbing, cover it all with a metal roof, and install the doors. The house is now move-in ready, and the only thing it’s missing is a good realtor, a better paint job, and some pictures on the walls.
>> This is a story of how incredible it is to build software with AI.
Why did I do this to myself?
Do you know how sometimes you sit on a toilet and you load a video on YouTube or your favorite social network, but you can’t play the video with sound on so you are missing out on all the details?! Wouldn’t it be nice if every video had subtitles?
I’ve always wanted an app that could generate subtitles for my video quickly and easily, without having to pay monthly fees to some service, and we now finally have the technology to do it, quickly and cheaply. So, I got curious, and just started to build.
What I ended up with was SubtitlesKing. It takes any video, generates a transcript, and burns subtitles on it. This is not a novel idea and many companies provide this as a service to podcasters, youtubers, influencers, and anyone that cares about engagement on their video content. I doubt any of them built the entire application in 48 hours.
Why should you care? This is going to absolutely be the future of software development. We are going to have thousands of people building software for themselves and then offering it for an inexpensive subscription to fellow builders and consumers alike. For large companies this is also the time to forget agile and to adapt to an entirely new rapid development framework. If you have an idea on Monday, you must have a fully functional application ready to test by Friday. If you don’t, someone else will.
So, SubtitlesKing.com development process (the agony, and the thrill)
Before I start, let me set the stage: I’m not a “professional” developer. I’ve been coding on and off for about 15 years, mostly teaching myself, mostly using Ruby on Rail and recently JavaScript. The real pros know their language of choice inside out and craft code with the precision of Michelangelo. I’m more of a “throw spaghetti at the wall and see what sticks” person. Pro developers would find a billion arguments against what I am about to tell you, but with the latest AI tools, my approach has gotten a lot easier, and a lot faster.
To start building the app, I quite honestly did not give it any thought, I just started building it. To help me, I used BoltAI (https://boltai.com/) to stream Grok (https://grok.com/), ChatGPT o1 reasoning model, and an AI coding editor called Cursor (https://www.cursor.com/).
I wanted two things at the end:
(1) to get the project done as quickly as possible, and
(2) to understand what I was doing so I could extend it, reuse the component, and to build more projects like this one, quickly and for a variety of ideas.
As an additional challenge to myself, I set out to write the whole application in Go, a programming language I’ve never tried before. In the end, I didn’t write a single line of code myself, and the AIs did all of the programming.
I was able to use AI to fly through all the steps without fear, and without rest.
I started working on SubtitlesKing by extracting subtitles from just one video, on my laptop, typing all the commands in the Terminal, one line at a time. When that worked, I figured out how to add the subtitles onto the video itself. When that worked, I moved the whole thing onto the server in the cloud, and stopped physically telling the computer to do each step via a command line… etc. I made hundreds of little changes, built on top of each other.
If I wanted to get a detailed understanding of some steps, like how servers work and what was the difference between Debian and Ubuntu, or what it meant to actually upload files to a remote location, I would ask Grok or ChatGPT a question and read through their answers. For pure code execution, I would just describe what I wanted in plain English to Cursor, and it would generate the code. I’d run it, check if it worked, and keep going like that for hours. This loop meant I could build in short increments, moving forward with every new functional step.
Here is what I learned in two days: How to setup a Debian server, how to user SSH and SFTP, how to read (and write) Go code, how to setup SSL certificates on your own machine, how NGINX works, how to use FFMPEG and Whisper, how to direct-upload to server instead of S3, and lastly, how to make LLMs give you the code you want, instead of the code they think you want.
With that, I was able to make the following:
A website to host SubtitlesKing.com
A web server to process the files
Ability to send files from a website to the server securely
Engine that can compress incoming files, transcribe them, add the transcript in the form of subtitles back to the original file, and send it back to the user
Token based authentication, no logins, usernames or passwords
At the end, I ended up with exactly what I was hoping to have, and it only took me a weekend!
The only downside of this process was my human inability to stay up all night. Despite how fast the AI could churn out code, I couldn’t keep up with it. I stayed up until 2a.m. several nights in a row, and the lack of sleep caught up with me. While my AI partner was ready to go 24/7, I needed to rest.
Using AI is like using a pulley system. You could attach a rope to the weight and pull on it, or you can add some pulleys in between and then you can handle a much greater load with exactly the same amount of force. You still have to do work, but you can do so much more!
Here’s an example of a video that I can take and subtitle in just a few minutes.
https://x.com/kirillzubovsky/status/1887681421323739270
Welcome to the future of software development. Remember agile, the cool new framework which enabled companies to move from a waterfall design pattern? Well, forget agile, or call this hyper agile if you will. If you’re working in a team at a larger company, these AI tools can speed up your “sprints” from weeks to mere days—or even hours. You could theoretically build a working prototype for a complex product in no time, handling the database, server, and front end on your own. The design might not be perfect at first, but for rapid testing and proof-of-concept, it’s a game changer.
The companies that embrace this new technology, do away with management hierarchies, and allow employees to experiment live, those companies will win by a massive margin. They will move and change so fast, their competitors will not be able to respond. By the time one company had a meeting about a potential approach, the other company had already run a hundred iterations. It’s like competing with a professional marathoner. If you practice by running 3 miles and they run 30, you’ve got no chance.
There’s a silver lining though. For now, humans still have to decide what to build, why it matters, and how to market it, as well as how to sell it. AI can help, but it’s not yet amazing at doing it on its own. In my case, I have plenty of ideas for improving my subtitle generator, but there’s also user feedback, business strategy, and countless other tasks that can’t (yet) be fully automated. But once again, AI can help.
Even writing this blog post, I was too tired to type everything out myself, so I dictated my thoughts and then used AI to organize them. It’s a glimpse of how work might look in the future—less typing, more conversation with AI, and faster iteration than ever before.
• My Goal: Take a video file, extract the transcript, and then recompile the video with subtitles.
• Old Approach: Setup multiple APIs, cloud services, and tricky pipelines. Very expensive just to get started (hundreds of dollars) in software and storage costs, plus endless hours of work.
• New Method: Using AI, build a prototype in five hours, then finish it off for a total cost of 48 hours and $29/month, soup to nuts.
Final Thoughts
Building this app in a weekend was a thrilling experience. AI let me do in 48 hours what might have taken weeks or months before. But the real challenge wasn’t the technology; it was keeping myself going without burning out. As AI accelerates development, we need to remember that the human body and mind still require careful pacing and rest. We will undoubtedly create more tools to improve our ability to juice the machines, but how fast and to what extent? It’s entirely possible that in the next 5 years we will make more progress in software than we have made in the last 25.
If you’re dabbling in software, or even if you’re a seasoned engineer, give these AI tools a try. Ignore the naysayers, and don’t worry about competition or lacking in skills. AIs can teach you anything about everything. Just don’t forget to take care of yourself in the process.
Happy coding (and happy sleeping)!