Guidance on AI coding tools
Table of contents
Abstract
Establish some guidance on using AI coding tools when contributing to the Carbon Language project. These tools have growing popularity and interest, and it would be good to have a clear and actively documented set of guidance for folks interested or already using them.
Problem
AI-based coding tools are wildly popular at this point and we should have active guidance about how and when to use them when contributing to Carbon rather than being reactive.
Background
Proposal
All submissions to Carbon need to follow our Contributor License Agreement (CLA), in which contributors agree that their contribution is an original work of authorship. This doesn’t prohibit the use of coding assistance tools, but what’s submitted does need to be a contributor’s original creation.
Carbon’s license was also selected specifically to maintain full compatibility between any contributions to Carbon and contributions to LLVM so that we can potentially move things between these projects easily. We want the same to be true regarding the use of AI-based coding tools, and so any contributions to Carbon should also abide by the guidance in the LLVM Developer Policy around AI generated code
This proposal updates our contributing documentation to contain both these points.
Rationale
- Community and culture
- Explicitly documenting what is and isn’t required when contributing to Carbon makes the project more open and welcoming to new contributors.
Alternatives considered
Place more restrictions around AI-based tool usage when contributing
We could consider adopting more restrictions on how AI-based tools can be used as part of contributing to Carbon. However, at this point we don’t have a compelling rationale for this.
There are real concerns around the quality of code output from AI-based tools in some cases. However, we should not rely on avoiding the tools to provide protection from low-quality code. Instead, as we already do, we have a system of code review, coding standards, and extensive testing to ensure high-quality contributions regardless of the tools used. The quality, correctness, and utility expectations of contributions are true regardless of which tools are used as part of authoring the contribution.
We have also seen abuse where large volumes of automatically generated “contributions” have been sent to projects, overwhelming their community. Again, this is never OK, regardless of which tools are used to achieve it. We hold the contributors responsible for using any and all tools responsibly and making useful and constructive contributions with them.
Last but not least, we also understand that there may contributors who do not wish to use these tools. Currently, we are not proposing any tools in the required workflow of the project.