With a little help from my "friend"

The last two major features I need to implement before the launch of V2 are the Community Creations page and the Backstory Editor for monsters. These two features are closely connected, and I wanted to improve their quality and user experience compared to V1.

For Community Creations, I needed the publication process to be moderated, and I wanted to give users the option to suggest images for monsters. For the Backstory Editor, I wanted it to be based on Markdown and to feel similar to an editor like Obsidian, paving the way for the Adventure Editor I want to build in the future.

Realistically, how long would it take me to implement all of this on my own, in my free time? Possibly another year, maybe more... and that's not acceptable.

A foolishly ambitious project

The last time I asked ChatGPT for advice and told it how long I had been working on this project, it was the first time I felt apprehension coming from a computer. "4 YEARS? For a D&D side project? Are you out of your mind?" is basically what it said, before firmly suggesting that I release what I already had and improve it later.

But I couldn't do that. I couldn't remove features I had already implemented in V1, and I needed to rebuild them in a way that would also support the OTHER features I wanted to add in the future. If I wanted to make all of this a reality and release it within a couple of months, I had to change my strategy, and I probably needed some help.

Over the last couple of months, I've been reading about what the developer community on X has been doing with AI, and how some people have been using it to create entire projects in a matter of days. So I decided to make a small investment, subscribe to an OpenAI Pro plan, and get almost unlimited access to GPT-5.6 Sol, one of the most advanced AI models available at the time of writing.

Document the documentation

Ok, so now I have an allegedly omniscient, unstoppable coding Terminator at my disposal, but how do I tell it what to do?

"Build me an Obsidian-like editor and make no mistakes" wouldn't work, because everything there was to know about how I wanted to build it existed only in my head, and I needed to transfer that knowledge to the AI. But first, I had to find a way to remain in control of the final result instead of letting it spit out code that I would have to fix later.

The first thing I... we? did was write down a plan describing how we were going to work together, which decisions my buddy Sol could make on its own, and which ones I would have to approve before they could be implemented.

We decided to build a generic "documentation framework" that could work for any type of project. It would define those boundaries, establish the rules for every step of the analysis and implementation of a new feature, and explain how each step should be documented so that, at any moment, another AI or a human could pick up the project and continue from where we left off.

Once we had this in place, Sol told me that I needed to write a detailed document explaining how I envisioned the Community Creations page, the publication process, and the Backstory Editor working, including how they would interact with each other. Once I had written everything down, Sol analyzed it, prepared a draft of the first analysis documents, and then a five-day-long survey began.

The five-day-long survey

For five days, from Sunday to Thursday, I spent all my free time answering hundreds of questions from Sol. These questions were designed to uncover exactly what I wanted, fill every gap in its knowledge of the project, and explore all the consequences of the decisions I was making, including many things I had never even considered before.

The complexity of the project finally started to become clear to me, and I could see no end to the questions pouring in. Was all of this worth it? Was the AI just trolling me at this point?

Finally, on the last day, I was asked to approve the completed analysis document, and the implementation of the first "milestone" could begin.

The first phase consisted of a collection of small utility functions and internal components, specifically chosen by Sol to determine whether the technology stack we were using was worth pursuing. Once that was out of the way, it began implementing the first prototype of the editor, and this prototype... sucked!

The Markdown Editor

The first version we built was a disaster. Sol decided that we needed toolbars EVERYWHERE, and every row behaved like a separate document. Whenever you moved the cursor inside one, two additional toolbars would appear above and below it.

It felt like a nightmare to use. I had clearly failed to communicate my vision properly, so we went back, defined the core principles of how everything should look, and started again.

The second version almost made me shed tears of joy... well, it still sucked and had dozens of bugs, but it showed me that we were on the right track and that what we were trying to build was actually possible!

This version still had one huge flaw: you couldn't select text across multiple rows. To make that work, Sol had to rewrite almost the entire thing once again.

The third version is the one you can see and test on the Editor and NPC Generator pages right now, and it finally feels like a real Markdown editor. It's not at the level of Obsidian or Typora yet, but it's incredible that we were able to build it in just a week and a half, and that it is already usable and functional.

-Ismael