I was very happy to see that, throughout the life of FEDERATE, the EU-funded coordination and support action bringing together the Software Defined Vehicle ecosystem, we were able to establish open collaboration as a success story. That success translates concretely into the Eclipse S-CORE project: during a recent high-level FEDERATE gathering, 이클립스 S-CORE was held up repeatedly, and generously, as proof that a “code first” approach to open source collaboration actually works in the automotive industry. It was a good moment, and I was genuinely glad to hear it.

But in the conversations that followed, someone raised a question I have not been able to stop thinking about since: what is the value of open source and code first, in a world where an AI agent (allegedly) could rewrite an entire codebase in a matter of days?

It is a fair question, and it deserves a real answer rather than a defensive one. So here it is.

A brief history of “code first”

When the Eclipse Software Defined Vehicle initiative launched in 2022, “code first” was a deliberately radical idea for the automotive industry. Instead of the usual top-down, specification-driven, consensus-before-commitment approach, we asked organisations to contribute actual working software: pieces of the puzzle, building blocks that others could use, extend, and build on. It was a bet that momentum comes from running code, not from documents describing code that might one day exist.

That bet paid off, and the narrative evolved with it. Building blocks became 이클립스 SDV 블루프린트: integrated, coherent sets of components rather than isolated contributions. Blueprints evolved into Eclipse S-CORE, the Safe Open Vehicle Core, which brought a serious, systematic focus on functional safety and an end-to-end platform for high performance ECUs. At each stage, the center of gravity shifted a little further from “Here is some code” toward “Here is a coherent, safety-relevant platform that an industry can actually build on.”

So when someone suggests that AI agents make “code first” obsolete because code is now abundant and cheap to generate, my instinct is to agree with half of the premise and disagree with the conclusion. Yes, code is becoming easier to produce. That was never really the point.

Three things AI agents do not replace

1. Safety is a process, not an output

In a safety-relevant domain like automotive software, writing code is a relatively small part of the job. The bulk of the effort goes into following a rigorous process: requirements traceability, verification and validation, documented evidence, review and audit trails that satisfy functional safety standards. An AI agent can accelerate pieces of this, certainly, generating code, drafting test cases, even flagging inconsistencies. But it does not replace the process itself, and in safety-critical systems, the process is most of the value.

This is, not coincidentally, exactly why we recently established the Eclipse Functional Safety Committee and formalised the Eclipse Foundation Functional Safety Process. The process gives projects a defined path to engage in practices consistent with recognised functional safety standards, operated in the open, transparent, vendor-neutral way the Eclipse Foundation operates everything else. It is, in effect, where pillar one and pillar two meet: the discipline a safety-relevant project must follow, backed by the governance and committee structure that makes that discipline credible to an auditor, a regulator, or a competitor sitting at the same table. An AI agent can help a team move faster inside that process. It cannot be the reason an outside party trusts the result.

2. Governance is not a side effect of code, it is a prerequisite for trust

This is a point I have made before, in the context of single-vendor open source models in automotive. Publishing code under a permissive license is necessary but not sufficient. What makes an ecosystem like Eclipse SDV valuable is vendor-neutral governance: a level playing field where competitors can collaborate on non-differentiating software without any single company controlling the roadmap, the access, or the rules of participation. That governance is not an afterthought, it is at the core of the services that the Eclipse Foundation provides and it is clearly defined in the Eclipse SDV Working Group Charter and in the Eclipse Foundation Working Group Operations Guide that spells out how decisions get made and who gets a seat at the table. That is a structural property, not a technical one. No AI agent writes proven governance. People, and institutions, do.

This also happens to align closely with where European policy is heading. As the conversation shifts toward technological sovereignty and independence, vendor-neutral, multi-stakeholder governance is not a nice-to-have, it is the whole point. Indeed the European Tech Sovereignty Package highlights Eclipse SDV as a model for industrial collaboration.

3. Community is the part that was never about code at all

This is, to me, the most important of the three, and the one I believe offers the greatest value that AI alone cannot provide. AI agents can write code, draft requirements, and even assist with safety documentation once a process is well defined. What they cannot do is align intent. They cannot provide thought leadership, build trust between competitors, or create the kind of shared understanding of what actually matters that emerges from people getting into a room, or a hackathon, or a community call, together.

We have seen this play out very concretely this year. Our community in Asia did not grow because we published more code or sent more newsletters, it grew because colleagues like Ansgar Lindwedel showed up on site, at meetups in Bengaluru and Tokyo (read the blog post here), hosted with partners like Elektrobit, and talked to people in the room. India’s first Eclipse SDV community meetup, held in Bengaluru in May 2026, is a direct result of that kind of presence, and the interest it created kept compounding from there. The same is true of the Eclipse SDV Hackathon: Chapter Four, held in Friedrichshafen from 6 to 8 October 2026, brought the whole community together in one place to build side by side, something a pull request alone could never do. The blog post by Christian is well worth reading, and it highlights the enthusiasm of the participants.

No matter how capable AI agents become, this part does not get automated away, because it was never a code production problem to begin with.

The irony worth sitting with

There is one more piece of this worth naming plainly. Every large language model capable of writing credible code today was trained, in significant part, on open source code. The commons that some now describe as being made obsolete by AI is, in fact, the foundation that made capable coding agents possible in the first place.

Eclipse Foundation projects use industry-friendly licenses precisely so that anyone can use the code however they see fit, and I think that is the right choice. But it does lead somewhere important: sustainability. If the assumption that “code is now free” discourages organisations from continuing to invest in and contribute to open source, the commons that AI agents depend on for training stops being refreshed. The quality of what these agents can produce is not independent of the ecosystem they were trained on. Starving that ecosystem does not make AI agents more capable. Eventually, it makes them worse.

So, what is the value of open source in an age when AI writes all the code?

Code was always the entry point, not the destination. “Code first” got organisations to stop arguing over specifications and start building together. What that process produced, over several years, was not just software, but a safety-relevant development discipline, a vendor-neutral governance model aligned with where Europe wants to go, and a community capable of aligning on what matters, together, faster than any single organisation could on its own.

AI agents are a genuine accelerant for the first of these. They have little to say about the second. And they cannot replicate the third at all.

That, as far as I am concerned, is the value of open source in the age of AI agents. Code was always the catalyst, never the end goal.

Sara Gallian Director, Software Defined Vehicle Program, Eclipse Foundation