Flarum development in the AI age

CyberGene

I’m a software engineer for a big company and we’ve been using AI-assisted coding for some time now and the impact has been tremendous. Especially with the latest Claude models, the level is ridiculous, to the point of being scary.

I used my personal Gemini 3 Pro subscription to implement a solution to the orphaned FoF Uploads and had a working solution in half an hour: a standalone Java program that uses a Flarum DB dump to scan through all the posts, connects to the AWS S3 storage to map all uploads and then prepares a pretty nice static HTML with analysis on all orphaned uploads with a thumbnail or a link for download and the ability for you to select those files that you need removed and it can then remove them from the storage, as well as from the forum DB “live” by using the Flarum REST API, or exporting a proper SQL statement that I can execute manually against the forum DB.

This all was ready basically in half an hour by me iteratively asking for different refinements and it was already knowing the Flarum API and DB structure from the information on the Internet. It’s a feature we’ve asked for a few years now with no prospect of it ever being created. And I got it from an AI agent by just asking and explaining what I needed.

With all that in mind, I’m wondering if the Flarum dev team is already utilizing AI-assisted coding? Because it can speed up things enormously. Especially since you’re just a few people with little free time to finalize the 2.0 release and no prospect of fixing all the open issues and features we’ve asked for years. Until now. What’s your opinion?

Anonymous77

Hope Flarum supports MCP soon. It'd be great to just use AI to write plugins instead of babysitting Composer's messy versioning and dependency trees. Highly extensible setups like Flarum are perfect for the AI era.

MichaelBelgium

As a senior dev of 10 years, I'm not really an AI fan here.

I use it a few times a month, every damn time i have to review and correct it. Which is exhausting and time consuming. The time and technical debt is humongous and not worth it to me. By the time you write a proper prompt with all the context AI needs ... I could've implemented a new feature already with the necessary security and testing (AI won't do this)

It's worse than mentoring a junior dev with actual programming knowledge - which is actually rewarding. But mentoring a junior with only AI slop knowledge? Ah hell naw, I'm going away as far as possible. Juniors are currently deprecating themselfs with AI usage. They can easily be replaced if they rely 100% on AI.

For the times i use AI, I do agree Claude is the best out there, gemini in second. But as Chatgpt was "the first", people only seem to know that one, and it's the most awful. For everything, not even coding. Which definitely doesn't help.

As for AI + Flarum you probably already noticed meta posts are AI generated, which i commented on https://discuss.flarum.org/d/38960-flarum-200-beta8-released-20-is-now-feature-complete/45
Core team most likely uses AI already. But making an MCP is pointless, you have docs made by devs with the knowledge, know-how and context about the whole project, Just read it. A lot of open source projects are AI slop now, nobody wants Flarum to become slop either.

CyberGene

MichaelBelgium I'm also a senior engineer with 26 years of experience in Java backend development. I’ve noticed a clear division among my colleagues regarding their enthusiasm for AI. Some, like me, are very enthusiastic about it, while others, like you, are not. It seems like this will be a topic of debate in the coming years. Regardless of our personal opinions, AI is here to stay, and there are strong indications that companies are adopting it extensively. This is evident from the fact that some of the world’s largest software companies are actively integrating AI into their dev lifecycle.

But avoiding AI in open-source projects is to me really baffling, especially on severely understaffed projects like Flarum. The 2.0 release took years, and many of the core developers left the project, we'll never know the actual reasons but frankly speaking I felt some tension between them. I wouldn't be surprised if Flarum 2.0 wouldn't have been possible if AI was not adopted.

BTW, I’ve been requesting certain features or updates to extensions for years, but they’ve been consistently ignored by the core devs and the extension devs. And then, recently, I downloaded Claude Code and implemented three extensions in just a few hours. Maybe they are not well implemented, feel free to review them. But it's better than nothing, right? Or if not, then we could introduce a badge indicating whether an extension has been created with AI-assistance. This would allow users to make informed decisions about which extensions to install, especially if they prioritize maintaining a sense of purity. I don’t mind this idea.

—
BTW, I believe this "puritanism" about using AI in open-source projects is an ideological dichotomy rather than a practical one. Despite our passionate debates about specific personal experiences that support either side. I am deeply convinced, as with many new technologies and aids in the past, people tend to choose emotionally, sticking to a "camp" and then rationalizing their experiences to their "religion". I might be a victim of this, you might be. We all could be. The future will tell 🙂

luceos

AI is going to be just another tool in the toolbox of the developer. To those who know how to wield it, it will be immensely valuable, to those that don't it might be an annoyance. Although I've seen and had to deal with the drawbacks of AI I am also starting to see the benefits on every level of how I do things. Coming out of a burnout, claude now helps me manage my todo, review the value gain of each item, persist my todo for papertrailing to just name a few things. Technically it helps draft tests, though not always great ones unless you are very specific, it helps me rationalize alternative implementations, although not always taking in the right reasoning.

A week ago I tried to create a cooking plan using AI for the week. It failed to add another batch of 500g of chicken to my shopping list. AI is great for a lot of things if you're open to it, but it cannot take away human supervision as this will lead to disaster one way or another.

Can AI be used to simplify development, but especially extension development, in Flarum? Yes, definitely. Will it increase the risk of vulnerabilities, data leaks or catastrophically malformed data or files in Flarum? Yes, also. Is it worth investigating whether tooling can be made to pursue this avenue, I personally think so.

CyberGene This is evident from the fact that some of the world’s largest software companies are actively integrating AI into their dev lifecycle.

You must have missed the news that Amazon forced developers to use AI assisted tools, then had a 13 hour outage. Now it's reinstating senior developers to review AI-code 😏

MichaelBelgium As for AI + Flarum you probably already noticed meta posts are AI generated, which i commented on https://discuss.flarum.org/d/38960-flarum-200-beta8-released-20-is-now-feature-complete/45
Core team most likely uses AI already.

I don't want to be the one to spoil your party 💃 , but using emoji's in release announcements has been part of Flarum's DNA since the earlier releases. For reference the announcement of v1.1.0 Platypus: https://discuss.flarum.org/d/29188-flarum-110-released.

CyberGene

luceos You must have missed the news that Amazon forced developers to use AI assisted tools, then had a 13 hour outage. Now it's reinstating senior developers to review AI-code 😏

There will be a lot of these stories until the things settle and the best practices are established. In our company AI-assisted coding is also semi-mandatory, to put it mildly, and I've had my fair share of initial enthusiasm leading to over-trust of what AI produces, hence failure. I can firmly say that AI is much better than me and much faster than me most of the time. But, occasionally it does some pretty stupid things, yet presents them with great confidence 😃 So, it requires some babysitting and getting used.

Yes, it's a tool and very handy at that. It's like a chainsaw: you can get seriously hurt using it, but if you know what you're doing, you'll be much more productive than using an axe. And BTW, you can get hurt with an axe too 😉 I mean, there's been so much bugs, vulnerabilities and various silly issues even in the best software prior to the AI-era. If something is AI-written doesn't necessarily mean it's of low quality, and vice versa.

The progress of the AI code assistants has been impressive. One year ago it was rather rudimentary and could only help in boilerplate code. But literally months ago with Opus 4.6 and Gemini 3.1 Pro the things got drastically better and with a bit of back and forth it can do complex software analysis even when thrown at tons of legacy spaghetti, and it can debug it, find issues, refactor, improve, optimize. It can now design and implement complete sub-systems, modules or whatever is the analogy in your field. If it keeps improving at that pace, I'll be really worried about our jobs. I hope I am wrong about that. But if you think of it, programming is not magic, it's not art. Yeah, a lot of people in our business prefer thinking of themselves as some special breed, doing something that cannot be replaced but the reality is coding and software design are highly formalized and deterministic.