Azhora III: The Lizeem War

As I mentioned in my last post on this game, I got kinda carried away with fantasies of an “everything game.” I am now focusing on the Lizeem War as a first coherent campaign and proof of concept, putting the more ambitious skill systems and broader cast of world characters on hold.

The idea is a fusion of two of my favorite kinds of games: 3D role-playing adventures and 4X grand strategy campaigns.

I want the player to be a hero in a larger, dynamic war that unfolds around them whether they intervene or not. Armies march, battles resolve, and territory changes hands. The player can influence those events, but the campaign has a life of its own. My longer-term idea is for the larger game to build toward an epic Ragnarök-like climax, where the player’s alliance must confront a powerful invading army. Developing that endgame is still a ways off, though.

Even finishing this one opening scenario — the Lizeem War — to my satisfaction leaves me with more design decisions and world building than I have the bandwidth for now. It has given me a clearer idea of what I want the game mechanics to be and what I am working toward, but to really do this game well I would need to make it my main project for at least a few months. For now, I am comfortable keeping it as a side project for fun and learning.

The development experience itself remains strange. Virtually all of my time goes toward design decisions and managing AI usage. I spend surprisingly little of it reading or writing code. In a sense, this is more fun, but it is also more overwhelming. Writing code by hand forces me to concentrate on the logic of a small problem. Working with agents lets me move quickly between ideas, which makes it very easy to keep expanding the project; however, I find myself tempted to add new features when I really should be refining existing ones.

I have also developed an insatiable desire for more tokens. Last month I ended up buying both ChatGPT’s and Claude’s $200 plans. Claude and Claude Code had become my preferred tools for a while, but since GPT-6 Astra arrived, I have turned to it for most of the development. Deciding how much usage to spend has become part of deciding what to build next.

Part of managing that usage is understanding what the different reasoning settings actually do. According to OpenAI’s reasoning guide, higher settings such as High, Extra High, and Max enable more reasoning on difficult problems, generally at the cost of more time and tokens. Ultra is distinct in that it uses additional agents to work on separate parts of a task in parallel. This means developers must ask ourselves before every prompt: how much computational effort does this particular problem actually need?

What continues to baffle me is how little of the implementation I seem to need to understand. I find myself wondering whether someone who couldn’t explain a for loop could have developed this. Perhaps having less sense of what should constrain them would even help.

This has raised the question in my mind: does knowing how to code make me better at this, or just less afraid to try?

I don’t know.

I do know that my ChatGPT Pro subscription expires today (my Claude Max subscription expired a week ago), and I don’t plan on renewing this month.

I need a break from coding agents.

I spent the last month trying to get as much out of these subscriptions as I could, and I think I derived a heck of a lot of value, but I also feel burnt out.

Beyond this game and my other software experiments, I found ChatGPT Astra in particular immensely helpful with giving me editorial feedback and even expertly typesetting my book. That has been my focus this last week, and I intend for my next post to be an announcement that the paperback is ready. Once that is done, it will be time for a significant break and recalibration as I prepare to look for a new job.