Experiences with Flyff Source + AI Coding Agents / Codex / Cursor / Claude Code etc.?
Hey everyone,
I’m curious if anyone here has already seriously worked on the Flyff source code using modern AI coding agents.
I don’t mean simply asking ChatGPT something like “How do I add feature XY?” and then manually copy-pasting snippets. I mean actually using tools like:
Codex Coding Agent
Cursor
Claude Code
GitHub Copilot Agent Mode
or similar agentic coding tools like gemini-cli
What I mean specifically is: opening the source locally, letting the agent understand the project context, analyze multiple files, suggest changes, fix bugs, refactor code, handle build errors, and so on.
I’m asking because the Flyff source is obviously not a small modern web project. It’s an older, grown C++/MFC/DirectX/server codebase with many dependencies, old patterns, sometimes cryptic structures, and usually a lot of custom systems on top. That’s exactly why I’m interested in how well AI can actually handle it today.
I’d be especially interested in the following points:
1. How well does the AI understand the Flyff source structure?
Does the agent understand things like WorldServer, DatabaseServer, AccountServer, Neuz, Common, the resource system, packet structures, etc.? Or does it quickly start hallucinating connections that don’t actually exist in the source?
2. Has anyone implemented real features with it?
For example systems like:
custom upgrade systems
new dungeons / instances
new UI windows
achievement systems
battle pass / daily rewards
pet / fashion / collection systems
new admin / GM tools
web API integrations
logging / anti-cheat / security fixes
Basically anything beyond simple text changes or small config edits.
3. How well does AI-assisted bug fixing work?
Especially for crashes, memory issues, packet problems, or weird client/server desyncs, I’d be interested to know whether an agent can actually help. Can it find the root cause if you provide logs, crash dumps, or compiler errors? Or is it more like: it sounds convincing, but in the end it’s still wrong?
4. What about refactoring?
Has anyone tried cleaning up old systems, restructuring code, or making certain parts easier to understand? I can imagine AI being very useful here, but also dangerous if it touches too many places at once and breaks something hidden.
5. Build setup / dependencies
How well do these agents handle old Visual Studio projects, libraries, include issues, and build errors? Can you actually work iteratively like: agent changes code → build fails → agent reads the errors → agent fixes the next issue? Or is the Flyff source too specific for that?
6. How do you control the changes?
Do you work with Git and review everything through commits/diffs? Do you only let the agent make small changes? Or do you give it larger tasks? I’d be interested in how you prevent the agent from messing up the source.
I’m not expecting AI to magically build an entire server by itself. I know you still need to understand what’s going on. But I do feel like these coding agents have become pretty powerful, especially when it comes to searching large codebases, understanding connections, writing boilerplate, and fixing errors step by step.
With Flyff this could be especially interesting, because many systems are old, messy, or poorly documented. At the same time, it’s also exactly the kind of source where AI can easily break things if it doesn’t really understand the architecture.
So my question is:
Has anyone here already seriously worked on the Flyff source using AI coding agents?
How was your experience?
What worked well?
What was complete nonsense?
Which tools did you use?
And would you say it’s already useful for productive development, or is it still more of a toy / experiment?
I’d really appreciate some real experiences, especially from people who didn’t just generate small snippets, but actually worked on the source code with these tools.
 |
