Lightning Talk: How To Responsibly and Effectively Contribute To Open Source Using... Tyler Helmuth

In this talk, Tyler Helmuth, a software engineer at Honeycomb, discusses the role of AI tools in open source contributions. He emphasizes that while AI can simplify the process of contributing by generating code and exploring repositories, it is important to approach contributions thoughtfully. Helmuth warns against submission issues such as large pull requests and excessive verbosity, which can overwhelm maintainers. He highlights the significance of building trust within the community through human interaction and engagement. Ultimately, he encourages contributors to leverage AI as a supportive tool while maintaining personal involvement and responsibility in the open source process.

Hello, my name is Tyler Helmuth. I am a software engineer at Honeycomb and while I started my career as a user of OpenTelemetry for the last 4 years, I've been lucky enough to be a maintainer and can support the project. I'm a maintainer of the OpenTelemetry Helm charts and the collector and I help out in other places as well and I also help support several open source

repositories at Honeycomb like refinery. As you all know, AI tools are everywhere. They've become a ubiquitous part of software development. These tools make it easier than ever to start getting to start understanding and getting involved in a new project. Even experienced contributors can benefit from AI tools. I'm using AI to write code every day now. It's simply becoming a part of our workflows like using an IDE.

And while it's easier than ever to contribute, these contributions are not always useful, quite the opposite in fact. Slop contributions, that is contributions with blatant mistakes, large change sets, or disappearing contributors steal time away from maintainers. Maintainer attention is a scarce resource. Maintainers have a limited amount of time each day to review contributions. Any time spent on slop contributions is time a maintainer could have spent reviewing

a helpful contribution. In my experience in an open source community, trust is the currency. The more a maintainer trusts you, the faster your contributions are going to be merged. New contributors need to build trust. This is built via their interactions, their code, and their commitment to the contribution. Only a human can build trust, a robot cannot. Open source is still a community of people. Robots need not

apply. Maintainers are not interested in in interacting with autonomous robots that are submitting contributions with zero human interaction. While these type of contributions might be able to solve a bug quickly, they create too much noise to be useful. So, we can immediately identify autonomous robot solutions as the wrong way to interact with an open source community. What are some other wrong ways? First is when you use

AI to create a contribution that is way too large. Large PRs have always been difficult for a community to handle. And since AI can create code so quickly, it can be tempting to put all of that code into a single PR. Don't do this. It is very hard to review. Instead, task your AIs to split up the change into multiple PRs. They are equally good at contributing

iterative development, especially when given a detailed plan, and can even chain iterations in different branches. Try to keep your PRs small and digestible. You want the reviewer to be able to quickly understand what you are doing. The next trap is verbosity. LLMs like to add lots of comments and make PRs with big descriptions. Overly verbose AI descriptions result in a lot of content that the maintainer has

to review before they even start reviewing your code. Instead, even if you generate all of your code with AI tools, it's best to open the PR yourself. Keep your description succinct, follow the PR templates, but most importantly, when a maintainer leaves comments, reply to those comments yourself. Again, open source is about people. Maintainers want to see that you're interested in the And finally, to re-reiterate, the final

trap is automation. While it seems efficient to automate every single interaction in an open source community, this is ultimately going to make you lose trust with that community. We've already seen several situations in OpenTelemetry where autonomous activity resulted in a can- accounts getting banned because it was too noisy. Instead, be present. Be the human behind the screen interacting with this online community. Be active in GitHub, Slack,

and the SIG meetings. Introduce yourself. Get to know maintainers. The more of the community knows you, the more trust you'll gain. So we've talked a lot about how not to use AI tools when interacting with an open source community. So, what are the right ways to use them? First, there has never been a better way to explore an open source repository than with AI. Using AI tools

to help explain the repository, how it works, what it does, how it's organized, what its contribution guide says, what its CI is, or what it makes what it make targets do is a great experience. Let AI help bring you up to speed on everything you can about this community and its repositories. The next answer is obvious. Use AI to write code. While AI tends to be an

awful communicator in open source, it is great at writing code. Just make sure that you are reviewing the code before submitting. That is still your responsibility. And finally, use AI tools for local reviews of your changes. Automated reviews are great for catching bugs ahead of time and will reduce the amount of things that a maintainer needs to Doing it locally instead of in the PR itself also

cuts down on the number of comments that a maintainer has to review before they can start looking at your In summary, the best way to think about AI tools is as tools. They are excellent tools, but poor contributors. Open source is a community that is always looking for new contributors. And right now, it's easier than ever to get started. No matter where you choose to contribute, we

look forward to engaging with you. And if you want to talk about these topics, if you feel passionate about about AI and contributing, uh you can find me at the Honeycomb booth or after the talk. Thank you.