I rebuilt Panda Wonderland on my own with AI, working with Claude Code. That earlier relaunch led to a question I've kept coming back to: how do I make this way of working sustainable beyond a single rebuild?
I want to spend more time deciding what the game should become, with less time lost rebuilding context and checking what changed across disconnected pieces.
First, build the foundation.
This time, I've been working solo with AI agents to design and develop my own game backend from the ground up, building on open-source projects. I call it Starward Game Services, or SGS. I then migrated my already-released commercial game, Panda Wonderland, onto it, alongside developing client integration and verification tools.
Designing and developing SGS itself has been the central work. Using my own released game gives me a real setting to examine the system, as part of the broader game-development loop I'm building toward with Starward.
Judgment still matters.
Working through this with Codex has reinforced something from the previous relaunch: clear direction and expert judgment still matter. I need to define what good looks like, review the result, and be able to follow what happened when something goes wrong.
That has led me deeper into verification, runtime tracking, and the handoff between people and agents. My goal is a workflow where what we learn from a running game can inform the next change, with evidence that a person can actually review.
A game to learn from.
Panda is becoming the place where I test those ideas in practice. This is still a work in progress, and I'm learning which parts help and which parts add unnecessary work.
The questions are practical: how do we reduce differences between what we intend and what an agent implements? What evidence lets us judge whether the work is correct? Which decisions belong with people? And how can evidence from a running game feed the next test or improvement?
I'll keep sharing the decisions, the mistakes, and the systems taking shape behind the game. Over time, I hope this becomes useful to other people building games with AI too.
Adapted from my published LinkedIn account ↗. This describes my own development work and ongoing experimentation; it does not announce a new store release.