Open source didn’t get less important because AI can write the code

Every model that writes your code is trained on code from open source projects. Which makes the question “is contributing to open source still worth it?” increasingly more relevant. In this blog post we’ll cover the opinions of Naor Peled to answer the question. Naor Peled is a software engineer at groundcover, a core maintainer on TypeORM, a maintainer of PR-Agent, and a contributor to Cyphernetes. He’s been maintaining projects since 2022. This post aims to answer that question in addition to covering what maintaining looks like now, what AI-generated PRs have done to his review queue, and the one filter he uses to separate a real contributor from a coding agent’s drive-by.

Q: How did you get into open source?

Naor started by fixing documentation for Vue and React and translated parts of it into Hebrew. In mid-2022 he joined the maintenance team of DynamoDB Toolbox, which was the first project he maintained. He was also a user of the library and had a list of things he wanted to improve, so he reached out to the author. The author was open to Naor co-maintaining by closing issues, reviewing PRs, that kind of thing. After that other open source contributions and maintainer opportunities came more easily and frequently.

Q: What’s a realistic first contribution for someone who’s never opened a PR on a project they don’t own?

“Start with the docs”, Naor shared. Look for typos, or for the thing that confused you as a consumer of that technology. After that, most projects label issues “good first issue” — maintainers mark those specifically for newcomers. Ask to be assigned, and once you are, you can usually get help from a maintainer. Being assigned gives you a reason to go investigate one area of the codebase, and from there, more significant contributions get easier. It’s still a good approach because it shows the community that you’re interested in learning about the technology. Finally, reading through docs or tackling “good first issue” is still a great way to learn. AI hasn't changed this because reviewers can tell the difference between a thoughtless vibe coded response with a poor fix and well reviewed and approved LLM assisted fix. We’ll dive into this a little more further down too.

Q: A maintainer can hand a small issue to an LLM in thirty seconds. Why keep labeling good first issues at all?

Time constraints still make labeling first good issues worth it still. Naor explained how he can offload a lot to an LLM, but he still doesn’t have time to validate everything and think about all the edge cases. And opening an issue for even a small task lets someone become part of the community, which is more significant to him than the fix itself. “I want the project to grow — more users, but also more people who are interested in maintaining it”, he explained.

He goes on to share that, if your maintenance team is one or two people, you don’t have time to go over many, many issues. So people outside the team who want to help are always a good thing. In fact, a few days ago on PR-Agent, someone asked to be assigned an issue, worked on it, and then wanted more tasks. To Naor, that’s a big win. The project gets more love, features get added, and he has another person reducing the load on my free time. While he could ask an LLM to do it, it’s not as valuable as creating community. Community will always be more powerful then one person using LLMs, especially when those developers can also leverage these tools. The benefit of community hasn’t diminished simply because we have tools that enable us to write code faster.

Q: What does maintaining TypeORM and Cyphernetes actually involve day to day? What is a contribution you’ve made that you are proud of?

Noar joined the TypeORM maintenance team about a year and a half ago, after using it for six or seven years. The author didn’t have time to maintain it day to day, so there were many open issues and PRs that never got addressed. Naor wanted to be part of that project and cut down the backlog. Now they have five or six very active maintainers. He runs the maintainer meeting every two weeks, review PRs, push features, and steer where he can. For Cyphernetes is the Kubernetes query language and Naor joined the project shortly after it started. Same shape of work: closing issues, adding features. Across every project I maintain now, it’s mostly reviewing PRs, maintaining docs, running maintainer meetings, and pushing things forward.

For TypeORM, one of the best contributions was less about the fix itself and more about the impact it had on the community. It a long-awaited fix that was half bug and half feature, and had been waiting on real attention for a very long time. Naor describes how he had the pain of that specific part of the library mysel and how he still shared the joy of the fix with the community. As contributing the fix alleviated the issues that a lot of other people had experienced too.

Q: A model can produce a plausible patch in seconds. Why is contributing still worth it?

Two reasons. The first is that LLMs are trained partly on open knowledge, so open source contributions actually improve them and steer, in some sense, how they think. And even when AI writes most of your code or all of it, you still have your judgment. You make the calls. This is the trade-off of one option, this is the trade-off of the other, and at the end of the day you say which one you’re going for.

Second, it gets you out of your comfort zone: a technology you don’t work on day to day, a language you don’t usually use. And you meet amazing people. It also challenges you as a dev because you’re forced to think about maintainability, scaling, performance. It challenges you to be responsible for more and focus on larger problems which devs might not get as much exposure to working on an engineering team at a large company.

Has the volume of AI-assisted PRs burned you out?

Naor admits that at some points AI-assisted PR burnout is real. He shares a common experience where he had a bounty on a TypeORM issue and got 20 or 30 PRs for it. Everyone tried to solve it the same exact way, and it was obvious the people behind them had just handed the LLM the issue and stopped there. Some of them didn’t even address the comments from the AI code review on their own PR. Going through all of them was a waste of time, and it was a bit sad, because the underlying issue was real and painful and it needed to be solved. This AI-assisted PR fatigue is real and new but it doesn’t undercut the importance of community. If anything it amplifies it.

When there’s a stream like that, Naor asks for help from the community. That’s what I really like about open source: it’s a community, so I can go to people who’ve contributed before or to my co-maintainers, and we split the load until it’s manageable.

Q:How do you tell a real contributor from a drive-by?

Naor shares that he uses an AI reviewer to comment on the PR automatically. Just as AI can be used to avoid critical thinking, it can also be used to filter out those who aren’t seriously interested in contributing. If someone submits an AI generated PR and doesn’t respond to the comments, it shows you whether the other person is taking their time and addressing feedback at all.

He’ll also ping contributors. He shares, “I’ve seen PRs where the person clearly opened it with a coding agent and left. I tag them, and if they don’t answer after a while, I close the PR. That keeps the mental load down — I don’t have to go over anything before I’ve seen the actual effort that went into it”.

Meanwhile when someone does put in the effort, they reply, they tag him, they tell him the rationale behind the way they did things. That really motivates him to go through the PR and validate it myself. Even if an agent did the task.The takeaway here is that it’s still easy to differentiate between slop and effort and effort still gets rewarded in open source. 

Q: What’s under-discussed about open source right now?

The AI innovation happening in the open source projects like OpenClaw, nanoclaw and Hermes Agent, and the specs around MCP and skills. They’re out there, and anyone can contribute to them. That makes it a really good time to get into open source, because you’re not only teaching an LLM. You’re steering how software is going to be written and used in the future. For this reason, Naor argues it's more important to contribute to open source now than ever.