The Importance of Domain Knowledge on the AI Era

About a month ago, I hit an important realization on the realities of using AI.

When I was at Xbox, as expected, we were on the token maxing stage of using AI for your work as much as possible. After the big improvement AI had in December 2025, we started taking it to new heights. I started experimenting on what I could do; this change was not only about using AI, it was about reimagining how work could look like with the new AI capabilities. After a few months I was doing what previously could take 3-5 engineers, in a fraction of the time.

I would first create a big specification document. It outlined what the work was about and a proposed split of work to parallelize as much as possible. I would go over a few iterations with my team to get it approved and check-in on the code repository, as the source of truth of what needed to be done. After that, I would use an orchestrator agent that spawns agents based on the tasks we outlined. With this setup the amount of things I was able to get done grew exponentially. I felt I could take any tasks, any project and kill it, thanks to the power I was harnessing with AI.

Fast-forward, I got laid off 🙃. And given my new set of skills and passion of wanting to make games, I said to myself:

“I could use AI the same way to make games as if I were a full team”.

The first week or two, I started figuring out the setup. Making sure AI could use the engine (via MCPs), sort out what AI subscription to pay now from my pocket, and creating context documents to have some foundation in how I want the agents to work.

Then, something I didn’t expect happened. When I started working on game ideas and building prototypes. I started with a very simple setup just 1 agent, 1 chat, 1 idea. Then, AI made something “that almost works” on the surface, but when I looked at the code in the engine…I knew it didn’t look right. It had poor composition, and I wasn’t able to navigate or understand it near as much.

I want it to make adjustments but directing it was feeling harder than usual…After bagging my head against the AI chat trying to make it change things over and over again, I got more frustrated and more lost. That’s were it hit me:

My instincts and knowledge of Software Engineering were telling me things were not right; but I lacked the gamedev knowledge to be able to direct AI effectively. Now even thinking about doing something like having an orchestrator agent helping me, would be useless. Agents will derail and I would not be able to tell them what’s wrong and what needs to be fixed.

“Learning medicine consists in part of learning the language of medicine.” - Daniel Kahneman ¹

I know Software Engineering, and programming is a big part of making games. But I didn’t have the gamedev language: The concept about what “Game Juice” is, what “Tweens” are, how gamedev programming differs from applications programming. All of this were things I still needed to learn!

AI makes us feel we can take into any project, any task and be able to “produce something”; but don’t fool yourself. To be able to do something professional, with good quality and cost-effective. The domain knowledge is paramount.

AI may know every language on the world now, but you don’t.

#Ai-or-Not-Ai