Software development is changing fundamentally right now. As AI coding assistants are getting better and better, developer productivity increases drastically. Everyone is using those assistants, so not using them on AntennaPod would mean that the competition has an unfair advantage. While active open source developers can apply to GitHub for free copilot access, this might not fit into all use-cases. Most other assistants cost money.
Our donation expenses policy does not cover this type of expense. Also, it doesn’t cover hardware costs for test devices (I spent about 250€ on Chromecast and wearos devices that I bought primarily for AntennaPod development. I don’t want to get reimbursed for those, but want to start the discussion on those as well).
I would like to propose the following new type of expenses that can be refunded:
Developer Tools
Active developers with a proven track record are eligible to apply for reimbursement of tools that they need for AntennaPod development. This includes AI subscriptions or (in rare cases) even test hardware.
Developers are eligible for applying for AI subscriptions if they contributed at least 6 substantial merged pull requests and have been consistenly active in the project for at least 3 months. Developers are eligible for applying for hardware reimbursement if they contributed at least 30 substantial merged pull requests and have been consistently active in the project for at least 2 years.
For now, I would not automate this decision and simply make a spending proposal on the forum for everyone interested. If this gets too much, we can always automate it later.
Of course, the fact that development gets AI assisted doesn’t mean the entire app will turn into a vibe-coded mess of unstable spaghetti code. Nothing changes about the code reviews and exhaustive testing that we already have. It does, however, highlight a problem we already had before AI: Writing code is relatively quick. Reviewing and maintaining it is what actually takes the time. So we will continue to have needs decision calls where we decide what features we want to accept and which ones we do not see fit into AntennaPod. And we continue to lack developers who want to work on reviewing other developer’s code.
I don’t disagree. Could I also ask that if you are allowing AI coding tools that in the repository you start creating AGENT.md and / or CLAUDE.md files that specify some of the rules you want the coding tools to follow? Of course that means you might have to start testing tools like Claude Code.
If you use Claude Code, you might consider adding a few .claude/hooks as well such as mandatory unit test running or a ratchet to make sure the unit test coverage doesn’t go down?
It’s hard to tell how much we should reimburse for AI tools. For example if I bought the $20/ month plan for Claude, how much of that will AntennaPod reimburse?
For every merged PR, AP will reimburse $20 or the usage of AI for the period of PR open to closed?
I’m just giving an example. Would be nice to make this simpler? 1 PR == 1 month of basic AI coding tool since it’s really hard to tell how many tokens were used for the PR
We already have an AGENTS.md file Of course we are not really looking for people 100% vibe-coding PRs that they don’t even understand themselves. We had that in the past, and code reviews were a pain.
My idea was to reimburse the full plan if you are an active AntennaPod developer. So you ask on the forum BEFORE you buy something and the community decides. Before the next month starts, you ask again and we check whether you contributed in the last month.
If this becomes too much noise, we can later change the process. I think it should be okay though, because there are not too many active developers.
Shouldn’t this be added to/specified in the donation expense policy as well?
For Expense proposal: Claude Code you went for a period of 3 months, which I think makes total sense. Maybe we should make that (or 2 months) the default?
I guess it just depends on how active the contributor is. If someone is rather new, maybe one month is better.
We have this sentence, so the “before” part should be covered:
Expenses should be proposed to and discussed with the community on AntennaPod’s forum. If the expenses have not been discussed and approved beforehand by the community, their reimbursements cannot be guaranteed.
I’m not completely serious here, but would you rather 100x your donation to sponsor a graphics card to contributors that can run open-source models locally (~10k€)? Running AI locally is much more expensive than using those cloud services
It doesn’t make much of a difference if Nvidia gets its money via one of these three (I only name those that came to mind first, there are of course others) or yourself directly.
AntennaPod doesn’t need to compete with other clients, whatever that means.
People leaving bad ratings on Google Play seem to disagree Also, google regularly makes changes in new Android versions that are a real pain to keep up with. We have waaay too few contributors working on the non-flashy things like bug fixing, and AI is surprisingly good at that when you give it access to the full source code of poorly documented dependencies.
In a way, it does have to. If people switch to some other app because AntennaPod moves too slowly for them, there are fewer people who contribute (and I’m less motivated to maintain if I know that spending 10+ hours a week of my free time only helps a handful of people).
I feel like this is not the right place to have an extensive debate about AI (LLM) use.
I have qualms many of the things that you said, starting with the negative reviews that are almost non-existent.
Furthermore, I’m happy to discuss these items further but my initial statement about giving these megacorps donor money still stands.
I think that, like the early Queen records that were emblazoned with NO SYNTHESISERS on the sleeves, that any app that proclaims itself to be AI FREE would be have an attractive selling point.
I think you have every right to decide whether you are going to donate based on your belief system.
If you are against AI in general, the money we as developers spend on our AI subscription will eventually go to one of the large tech companies. The world has changed for developers, we are using these tools to help us.
If you are against the money going to any high tech companies, I guess you can proposal a another funding budget for operation costs and donate only to that budget.
Belief system? I’m not sure where you’re going with this.
I’m not against AI in general.
I am against LLM in general unless some ethically trained LLM shows up. (Is that even possible?)
Only then could I maybe be ok with my donation being used for a probabilistic software manipulator subscription.
Then again I’d rather want to see the money used for hardware running such software but soon enough nobody will be able to buy hardware because of the skyrocketing costs of RAM, SSDs and hard drives.
This reads like you’re taking offence by the comment about belief systems? I think it’s a totally legit comment:
There are facts and there are opinions (which are hopefully based on one fact or another). One’s combined set of opinions could be called ‘belief system’ or ‘world view’, which will impact the decisions you take (e.g. around donations).
I’ll come out and say that my world view is that the way we have set up capitalism is f*ed up because it promotes innovation at all cost (including climate change, human suffering and invasion of privacy). At the same time I am interested in technical innovations even if they don’t necessarily have a clear world-benefit. When it comes to ‘ethically trained’ I care less about the thing we call copyright and more about the other (aforementioned) aspects.
My belief system is internally clashing here. Luckily for me, in the context of AntennaPod it doesn’t matter that it’s clashing as I’m donating my time and not money anyway.
(Let’s not suggest that offence should be taken by someone’s comment, and when you do take offence be clear about it.)
I’m not donating but agree with you on what to spend money on. But I’d assess the situation differently: already now nobody can buy hardware because of costs, especially hardware capable enough to run an LLM locally. Consequently: in the short term paying for SAAS is more realistic but hopefully that balance will tip around 2028/2029 when prices hopefully come down again.
I don’t think that for AntennaPod it’s realistic to have different ‘Jars’. I mean - we could set up different budgets in OpenCollective and allow folks to donate to those specific budgets, but with our budget I think it’d put unnecessary constraints on how money is spent.
Of course we are not really looking for people 100% vibe-coding PRs that they don’t even understand themselves. We had that in the past, and code reviews were a pain.
Also, google regularly makes changes in new Android versions that are a real pain to keep up with. We have waaay too few contributors working on the non-flashy things like bug fixing, and AI is surprisingly good at that when you give it access to the full source code of poorly documented dependencies.