Hi! Based on the results of the discussion, I've made PR https://github.com/apache/cloudberry/pull/1740
Yes, it was generated using various models - Qwen-3, Claude 4.6, GPT 5.5. The AI_POLICY.md is partly generated, partly written by me. AGENTS.md is fully generated. Please review it ) On Tue, Apr 7, 2026 at 5:30 AM Max Yang <[email protected]> wrote: > Thanks for raising this topic — it's timely and important. > > I fully agree that AI-generated code is already part of our daily > workflow, and pretending otherwise doesn't help anyone. Rather than > discouraging AI usage, I think we should embrace it with clear > guidelines. Here are my thoughts: > > 1. AI is a productivity tool, not a replacement for accountability > > 1. Just like IDEs, linters, or Stack Overflow before it — AI is a > tool. The developer who submits the code is still fully responsible > for its correctness, security, and compliance. This should be the > cornerstone of any policy we adopt. > 2. We need a developer-friendly AI policy > > 2. I agree that the ASF rules are hard to parse for most contributors. > I'd suggest we create a concise, practical document (similar to > ClickHouse's AI_POLICY.md) that covers: > > - Disclosure: Contributors should indicate when AI tools were used > in substantial code generation (e.g., a simple tag in the PR > description). > - Review standard: AI-generated code must meet the exact same > review bar as human-written code — no shortcuts. > - Licensing awareness: Contributors must ensure AI-generated code > doesn't introduce license-incompatible snippets. This is > especially critical for ASF projects. > - No AI-generated code in security-sensitive areas without extra > scrutiny: Crypto, authentication, access control, etc. deserve > additional human review. > - Testing requirement: AI-generated code should come with > corresponding tests. If the AI wrote the code, the human should > at least write (or carefully verify) the tests. > 3. Practical suggestions > > - Add a checkbox in our PR template: "This PR contains AI-assisted > code generation: Yes / No" > - Create a short AI_POLICY.md in our repo — written in plain > language, not legalese > - Periodically share best practices on how to use AI agents > effectively within our project (prompt engineering tips, common > pitfalls, etc.) > > The goal should not be to create barriers, but to make it easy for > contributors to do the right thing. A clear, simple policy actually > encourages responsible AI usage rather than pushing it underground. > > What do others think? > > Best regards, Max Yang > > > On Mon, Apr 6, 2026 at 4:54 PM Leonid Borchuk <[email protected]> > wrote: > > > Hi > > > > I recently read an article by the Clickhouse team about using AI to > develop > > database kernel/infrastructure code > > https://clickhouse.com/blog/agentic-coding and understood that the > > everything that has been said applies to our project too. > > > > Really, we can't deny that AI-generated code is here. We see it in the PR > > submitted from developers involved in a project. We use AI-agents > ourselves > > to write texts/code/responses and be more productive. > > > > So, my question is: since this is a new reality, perhaps it would be > useful > > for all project contributors to know the correct style of using AI > agents. > > > > Something like Clickhouse AI policy > > https://github.com/ClickHouse/ClickHouse/blob/master/AI_POLICY.md > > > > We also have a set of ASF rules > > https://www.apache.org/legal/generative-tooling.html. But they are > > difficult for a simple developer like me to understand (it contains a lot > > of legal terms). We could say the same but in a more developer-friendly > > manner and so avoid embarrassment. It is simpler to say that I am not > using > > AI agents, than to decide if it is legal and approved by society or not. > > > > WBW, Leonid > > >
