AI 시대: '3C' 프레임워크로 오픈소스 멘토링 재고하기

'Eternal September' in open source refers to a continuous influx of new contributors, akin to the perpetual stream of new users Usenet experienced after AOL opened access in September 1993. Traditionally, this influx strained social systems for trust and mentorship. In the AI era, this phenomenon is exacerbated because AI tools dramatically lower the cost of creating plausible-looking contributions. This means maintainers face an unprecedented volume of pull requests that appear well-crafted but often lack deep understanding or context from the contributor, making it harder to discern genuine investment and increasing the burden on reviewers. It challenges the established mechanisms for building trust and integrating newcomers into the community.

Mentorship is the lifeblood of open-source communities because it's how knowledge is transferred, skills are developed, and communities scale. A good mentor doesn't just add one contributor; they enable that contributor to eventually mentor others, creating a powerful 'multiplier effect.' This ensures the project's longevity and health. However, mentorship is currently at risk because maintainers are burning out. The sheer volume of AI-assisted, yet often context-lacking, pull requests means they spend excessive time debugging or providing feedback for contributions that don't reflect true understanding or commitment. If maintainers can't strategically invest their limited time, the mentorship pipeline breaks down, jeopardizing the community's ability to grow and sustain itself in the long run.

The '3 Cs' framework—Comprehension, Context, and Continuity—provides a strategic filter for maintainers to decide where to invest their mentorship energy. **Comprehension** assesses if a contributor truly understands the problem and their proposed solution, often checked by requiring an issue discussion before a pull request. **Context** refers to whether the contributor provides sufficient information for a thorough review, including linking to issues, explaining trade-offs, and disclosing AI usage, potentially via an 'AGENTS.md' file. **Continuity** is the ultimate filter, focusing on whether a contributor consistently engages, responds thoughtfully to feedback, and keeps coming back to contribute. This last C is key for identifying individuals worthy of deeper mentorship.

Disclosing AI use in contributions provides critical context for reviewers, allowing them to calibrate their review approach. When a maintainer knows a pull request was AI-assisted, they understand that the code might be syntactically correct and follow conventions, but the contributor's understanding of the underlying problem or trade-offs might be limited. This enables the reviewer to ask more targeted clarifying questions, focus on assessing the contributor's comprehension rather than just the code's quality, and guide them towards deeper learning. Policies like those by ROOST or Fedora for AI disclosure help foster transparency and manage expectations, ensuring that reviews are more effective and less time-consuming for maintainers.

'AGENTS.md' is a file that provides instructions for AI coding agents, functioning similarly to a `robots.txt` file but for AI tools like GitHub Copilot or other AI assistants. Projects like scikit-learn, Goose, and Processing use 'AGENTS.md' to specify guidelines for AI agents, such as ensuring they follow project contribution norms, check if an issue is assigned before generating code, or adhere to specific stylistic conventions. This mechanism helps maintainers by shifting some of the burden of gathering necessary context onto the contributor's AI tools. By setting expectations for AI-generated contributions upfront, 'AGENTS.md' can reduce noise, improve the quality of initial submissions, and streamline the review process for human maintainers.

Maintainers can apply the '3 Cs' by implementing clear guidelines and watching for specific behaviors. For **Comprehension**, they can require contributors to open an issue and get approval *before* submitting a pull request, ensuring an initial understanding. For **Context**, they can ask for specific review information like issue links, trade-off explanations, and AI disclosure (perhaps via an 'AGENTS.md' file). For **Continuity**, maintainers should initially offer limited mentorship, such as a teachable moment in a pull request review. Only if the contributor responds thoughtfully and *keeps coming back* to engage should deeper mentorship, like pairing on tasks or offering commit access, be considered. This strategic filtering protects maintainers' valuable time and focuses their energy on genuinely committed individuals, preventing burnout.

The 'multiplier effect' in open-source mentorship describes how one well-mentored contributor can eventually become a mentor themselves, teaching others, and thus multiplying the maintainer's initial investment. This exponential growth is vital for scaling open-source communities. The '3 Cs' framework helps maintain this effect by ensuring that mentorship resources are directed efficiently. By focusing on contributors who demonstrate Comprehension, provide Context, and show Continuity, maintainers invest in individuals most likely to become future leaders and mentors. This strategic approach prevents burnout from endless 'drive-by' contributions, allowing maintainers to nurture a core group of committed individuals who will perpetuate the knowledge transfer and community growth, thereby sustaining the multiplier effect even in the face of AI-driven changes.