Why My AI Project Hasn't Shipped After Twenty Days
- No.
- 192
- Date
- Read
- 3 min
- Words
- 580
For more than a month now, I've been working on an AI project. On May 5, I put out a 1.3 demo. The framework and usability were pretty much there.
But today, twenty days later, the official release still hasn't landed. It doesn't look very efficient, especially since AI is doing all of the programming.
Today I did a review and went over why the official release hasn't landed in twenty days.
First, for more than a week out of those twenty days, the project was completely stalled, just like my Official Account, which had no updates at all. Only after that did I start improving it further.
Second, while building, I ignored the complete framework in my demo system and made the whole process and system even more complicated, so what I built always felt a bit off to me.
Take design. From the start until now I've changed the design three times, because I never had a clear design direction or style in mind. So along the way, besides improving the app, I redid the design twice.
As for the framework, the system framework in the original 1.3 demo was already very solid and clear. I'd also been using it in real work for a while, and overall there were no problems. It only needed small tweaks to details.
The small tweaks I'm talking about were really about subtracting from what I had. But during development, I kept adding. I added two or three extra steps to the process, and I neglected the most important part of the product's positioning.
In short, after the demo was out, when I got ready to optimize it further, I hadn't settled on a clear approach beforehand. The design had no clear direction, and the framework was being built out as I went along.
So just two days ago I still thought two more days would get the first official release out. But today I worked on it all day and overturned two things.
- The current process design
- The current app design
Those two amount to a major overhaul of the whole app. I went retro on the process: starting from the original demo, I reworked the standard process this system should really have, with no redundancy. And I completely overturned the design I'd just settled on and started the interface design again.
The problem is that over the last few days, my Codex and Claude quotas are mostly used up, and I don't know if what's left will last until the first official release lands.
From this, I drew a few lessons:
- Software development needs order. As someone who knows nothing about AI, programming, or development, this is what I've learned through practice: split your needs into modules, and let the modules build on each other without looking messy. That's order, and an app built module by module this way feels the most comfortable.
- At the very start, you must be clear on what the core selling point of your software is. What problem does it solve? What's the biggest difference between it and other products in the same category on the market? You have to know that. I think good software is a good idea, and idea first holds true in many parts of life. This experience isn't necessarily a bad thing. They say you learn by doing, and for someone like me who knows nothing about programming and development, this kind of development experience actually helps my thinking more.