AI use for development?

[Chud]

Out of curiosity, how much are the active maintainers using AI here these days?

Quite a few years ago I stuck my toe in the water for helping out with bug fixes, to-do items in the code, etc. but it soon became evident that my available time combined with the level of my C chops meant that I wasn't going to be much help. More recently though I've been using Claude to help write code for various small things at work, and I've been impressed with its results (though they aren't good enough not to require oversight yet).

[agoodman]

as the borg guy (I think "maintainer" is too strong a word for me... I am not in the same sort of category as Nick or Eric) I can say that I don't use AI in coding the borg at all. I probably should but I am stubborn and obstinate plus I find hand coding more fun.

[Nick]

That goes pretty much exactly for me too

[JoseMachete]

My very utterly humble two cents as a variant dev, no AI for me. Just not my thing. Might be a tool I'm missing but I don't care.

[chem]

would the developers object if, at the official github, people contributed to analysis of existing issues, submission of new issues, or PRs using frontier AI models such as opus 4.8 or gpt 5.5? have you considered adding an AGENTS.md file with specific instructions for agents working with the repo? it could be similar, in part, to CONTRIBUTING.md, but with agent-specific instructions, or CONTRIBUTING.md could add a section relating to how you want agents to work with the codebase.

I just want to see angband keep going another few decades and it seems ai-assisted coding will inevitably impact this project too. may as well decide how to address it. if AI agents working with the repo is forbidden, that's worth stating too.

[backwardsEric]

I have three concerns about agents making more or less unsupervised contributions to the GitHub repository:

• If it turns into a flood of comments or pull requests, then that will exhaust the real humans who are active there.
• https://github.com/hengband/hengband has had automated reviews of pull requests for a while now, and I have not found that to be particularly helpful. I remain uncertain about how to meaningfully respond to a bot's review and, in most cases, that means I do not respond and wait for one of Hengband's human contributors to chime in. If other people act the same way in response to agents making comments or pull requests on Angband's repository, the agents will not be helping the conversation.
• For patches or pull requests, I do not know how copyright law treats non-human authors. Implicitly, the author of a patch or pull request is agreeing to have the work distributed under the terms of Angband's license. Does an agent understand that license? Can it be said to have legally agreed to the terms? Could it render a deposition about how the work was authored and what other sources were consulted?

Given those concerns, I would lean towards forbidding agents.

[Nick]

I'd add that Angband has not traditionally reacted to new trends hastily, I think we can afford to wait a bit and see how this all plays out.

[getter77]

Needless harm for nebulous, transient good for leveraging AI on Roguelikes(exactly one known project leveraging it which has managed any degree of relative longevity and forward momentum) on the whole, let alone Angband, let alone games in general. Forsaking the human spirit's touch while not even getting a Ghost In The Machine out of it, on top of it likely being a poison pill for posterity in terms of creating an utterly cursed spaghetti situation worse than what ultimately brought Portralis low.

[dos350]

in my observation; ai is quite handy on the angband codebase; noticed hengband got a agents.md; but dont know how hard they use it but yeah.

ai is a very powerful tool for angband development; a force multiplier

like me i wrestled the borg of 4.2.6 to run with oldclass.txt ~ maybe cost like a dollar worth of token and yeah its good

~~~~~ EDIT

going to ramble here a bit

a few more observations its like in my amateur opinion, ai can be utilized to very much do some heavy lifting in the angband code
e.g :
ports - ai could prettymuch one shot a port at least in terms of getting it running; (im thinkin sdl3)
cleaning up legacy patterns and doing targeted or broad refactors
and yeah prettymuch any general task of angband dev can be sped up by using ai;

tips for approaching it:

make some docs first and point him to it
if u are doing a targeted thing that is an in depth change / feature, ask for an analysis and docs first, this will help the ai work more efficiently and u can save the docs for later use;
dont be afraid to try things out!

angband already provides the scripts and etc for builds and tests and ai can leverage the history of git and etc to refer to certain things if needed.


i dont mean to go insane but yeah to quote locutus; resistance is futile

~~~

very much thanks to all angband devs for the work and care for this classic game for so many years!

[agoodman]

Interesting. The new VStudio has GitHub Copilot built in so I have been using AI a bit more. I find when I ask it things it depends on the category of what I ask. it frequently generates bad code so asking "make some code to do xxx" is not great but asking about bugs makes it so it can pinpoint some issues well. For example when there was a complaint that a new build slowed the borg significantly I asked it to analyze both and tell me what would help. It pointed out 4 things, the first two were good (both of which I had already found but, still, good on it for finding the right thing) and two were less helpful.
The AI also does predictive typing and that is helpful when I want to change something refactoring doesn't handle and it lets me just tab through "ya, here too change the formula to xxx".