Mark is posting a top level post that covers our thinking in more detail! Basically: we don't require anyone use it for development and we don't forbid anyone using it for development. We expect that if developers who are not us use it, they use it as an assistive tool at the steps the current tools are good/reliable at doing, to help make their process more efficient, and that the pull requests they submit have been human-reviewed: we would prefer not to get agentic or vibecoded submissions. Any artwork (mood themes, style images, etc) must be the work of a human artist, not generative AI.
Basically: the current tools are good at making it easier to do a lot of the things that are a significant challenge in a codebase of our size, age, and complexity, and the responsible use of tools like Claude and Copilot by an experienced developer, to do things like point you at the likely source of a problem, do first-pass refactoring that can serve as a structure that you start from, etc, is such a significant timesaver that it clears out a ton of roadblocks that historically have been some of the things we struggle most with in terms of development capacity: not using it when it's called for would mean continuing the same slow development process that has frustrated people badly in the past decade or so. (As well as frustrating the developers!) But we aren't letting it roam freely and we aren't letting it write code by itself or make changes without significant human supervision. We expect that if people use it, they use it as one tool of many, as a part of their process, and the final PR or commit is ultimately the product of human work and human judgement, no matter what tools were involved.
no subject
Mark is posting a top level post that covers our thinking in more detail! Basically: we don't require anyone use it for development and we don't forbid anyone using it for development. We expect that if developers who are not us use it, they use it as an assistive tool at the steps the current tools are good/reliable at doing, to help make their process more efficient, and that the pull requests they submit have been human-reviewed: we would prefer not to get agentic or vibecoded submissions. Any artwork (mood themes, style images, etc) must be the work of a human artist, not generative AI.
Basically: the current tools are good at making it easier to do a lot of the things that are a significant challenge in a codebase of our size, age, and complexity, and the responsible use of tools like Claude and Copilot by an experienced developer, to do things like point you at the likely source of a problem, do first-pass refactoring that can serve as a structure that you start from, etc, is such a significant timesaver that it clears out a ton of roadblocks that historically have been some of the things we struggle most with in terms of development capacity: not using it when it's called for would mean continuing the same slow development process that has frustrated people badly in the past decade or so. (As well as frustrating the developers!) But we aren't letting it roam freely and we aren't letting it write code by itself or make changes without significant human supervision. We expect that if people use it, they use it as one tool of many, as a part of their process, and the final PR or commit is ultimately the product of human work and human judgement, no matter what tools were involved.