Hello,
I did warn everyone on my first official day, standing on stage at Flock last year, that there were hard conversations coming. AI is going to be several of those hard conversations.
Here is why I’m not currently concerned about reputation damage in May 2026. Last month, the Linux kernel development community adopted an AI code generation policy that explicitly allows for the use of AI code generation. It compares well with the text of the Fedora policy, though they have more details in what should be in the Assisted-by tag to help with data collecting about which models are being used. As a project we’ll probably want to learn from their policy so we can do our own data collection concerning model use as we start seeing assisted by tag use in our project contributions.
I noticed you framed this concern differently than Fabio did. I was replying to Fabio’s specific concern about reputational damage. You’ve stated a wider concern about whether I’ve fully considered reputational impact. I feel like I also need to speak to that reframing. I have considered the impact, both the downside and the upside. I actually think there is net positive reputational impact in this approach. We are creating a space to be in conversation with the segment of the open source contributor ecosystem that values this technology as a development tool. This will help us gain influence with a segment of the community we aren’t well-equipped to reach at present. I know it’s a delicate situation, but I think we are threading the needle here. We avoid putting the AI tooling front and center with the users who don’t want it, but we build an output where it’s the entire focus, so we can get that segment of the community on board with working with us to bring this technology into a better and more ethical future.
As I noted in the second reply to Fabio, there is nuance in how the output from this initiative should be promoted. I’ll reiterate, don’t assume that this will be promoted in the same way Workstation or KDE Plasma Editions are promoted if this ends up being an Edition. Even if this ends up being a spin, I will end up needing to promote the output to specific audiences, because I view it as an important strategic bridge into a segment of the open source community that we have to be in partnership with in order to avoid the worst possible future.
Let me paint you a picture of the future… what I see 2028 looking like.
I’m looking at a future where it’s conceivable that the majority of commits into the Linux kernel upstream will have AI assisted-by tags in a couple of years. There are some people out there who think we’ll hit that mark much sooner than that, but I’m thinking 2 years. And where the Linux kernel goes, so does the bulk of the critical path ecosystem that makes the OS work. Yes, sure, there will be some projects where AI is not allowed at all. But I fully expect the bulk of the critical path components we build an operating system from will have AI-assisted code in it and assisted development will be normalized as an open source linux ecosystem development activity upstream of this project.
What’s the best possible version of that future?
It is an open question whether any of those commits will be done with open AI models that run on hardware contributors control — or if all will be by the large centralized AI models via APIs and pay-as-you-go token use. I believe that Fedora can help make sure it’s not the worst of those two futures.
In this future, users and contributors who have deeply held philosophical concerns about AI are going to tie themselves up in knots, knots of the Gordian variety. Are those people going to fork the Linux kernel and build an entire ecosystem devoid of AI assisted development? Are they going to jump to one of the BSDs that has taken a hardline stance? I don’t know. But I am pretty sure there is no AI-less future for the development ecosystem on which the Fedora Project currently depends. I think it’ll be hypocritical for any Linux distribution in 2 years to take a hard line stance on AI as a development tool. And I don’t see Fedora jumping to the NetBSD kernel.
Like all Gordian knots, it requires bold action. We have to do something. The question becomes what actions are impactful and achievable?
So what is actually achievable in the future? I think Fedora needs to be in the conversation about what exactly AI assisted development looks like broadly for the entire linux development ecosystem. We need to be pushing towards better, more ethical model use by offering a developer environment that, as a long term objective, includes making it easy to work with local models. We need to help the local models compete with the large centralized models for developer workloads.
Will small, more-open models be able to compete? I don’t know. But this space is under active development.. and as the big models seem to be hitting the laws of diminishing returns now. Maybe there is a chance the smaller models can catch up. The people that probably have the best idea about that are working in the AI space right now, and I need to build a bridge to them. This objective is one of those possible bridges. I’d like to not burn it, before we build it.
If we aren’t there trying to push towards the future with the most ethical AI development workflows that are achievable… we risk getting the worst possible future by default. I don’t think there is a future available devoid of AI-assisted development as a normalized workflow in the critical upstream ecosystem that we depend on. We can have the most impact by trying to figure out how to provide the most ethical version of this technology and put it into the hands of developers to choose to use.
This isn’t fun. This is a hard conversation. I’m not dismissing anyone’s concerns. In fact I probably agree with most of the concerns anyone could think to express to me about AI use. I live in the region of the US, Northern Virginia, that has the highest density of data centers. Both power and water usage are real concerns that I’m directly impacted by now that I no longer live in Alaska. There is arguably a drop in the water table regionally because of these things. The path we are on isn’t sustainable. Telling people to stop using this technology isn’t going to work. Offering people a more ethical version, which they can actively contribute to making better, may help.
I’m not dismissing the concerns. If I’m doing anything, I’m disagreeing about what to do with those feelings. I don’t think we can just ignore the technology or wish it away. We may be able to address some of the concerns via government legislation. But I think, as technologists, it’s incumbent on us to provide a version of this non-deterministic technology that is better in the ways we care about, so that our version is what open source developers choose to use when they choose to do assisted development.
That may mean getting our hands a little dirty at the start, so we can build the bridge into the segment of the open source community that sees value in the technology. This bridge should allow us work to make incremental progress towards an achievable ethical future. We do so knowing there is a gap between what is achievable and what is best as seen from where we stand right now. The gap may be too wide for some. And if you’re in that group, all I ask is you give us the space to walk towards what is achievable. By doing so, we might see the gap close because each step changes our vantage point.