[LFX] Define localization team guidance for Gen AI use and human review

[apullo777]

What would you like to be added?

Propose a lightweight instruction.md template that localization teams can adopt to guide their use of Gen AI in translation workflows.

The goal is to help localization teams define practical guidance for AI-assisted drafting, including:

• when AI use is allowed
• what must remain human-reviewed
• terminology and tone expectations
• suggested prompting practices that reduce reviewer burden and consistency issues

Why is this needed?

Different localization teams may increasingly use Gen AI under their own policies, but without lightweight team-level guidance, usage can become inconsistent and create extra review overhead. A simple template could give teams a reusable starting point while still allowing each team to adapt it to its own workflow and standards.

Proposed approach

Start with a lightweight instruction.md template for localization teams, rather than a modular framework or CLI. The template could cover:

• purpose and scope of AI use
• allowed and disallowed usage
• human review requirements
• terminology and style expectations
• prompt starter examples
• notes on safe and appropriate usage

PoC

Test this approach with a localization team that already has clearer policies and practices, such as the Korean localization team, and use that PoC to derive best practices before considering broader standardization.

/area localization

[k8s-ci-robot]

This issue is currently awaiting triage.

SIG Docs takes a lead on issue triage for this website, but any Kubernetes member can accept issues by applying the triage/accepted label.

The triage/accepted label can be added by org members by writing /triage accepted in a comment.

Details

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository.

[lmktfy]

I'm happy to have this work happen, but I don't understand it—after reading the issue, I don't think I could work on this myself, if I were asked.

Just something to consider.

[seokho-son]

Hi @apullo777,
Thanks for bringing up this issue! :)

Creating something like a language-specific (yet still broadly applicable across localization) instruction.md, intended for localization teams rather than individual contributors, does seem like a meaningful step forward, especially when considering future scenarios where teams leverage Gen AI based on their own policies.

However, the proposal in this issue feels somewhat overengineered at this stage, and a CLI may not be necessary yet.
A simpler approach might be more practical, such as providing a lightweight instruction.md template that teams can easily adopt to guide their use of Gen AI.

It may also be beneficial to start with a PoC in a specific language team that already has established policies and guidelines, for example the Korean localization team. From there, we could derive best practices before expanding into a more structured or modular framework.

[apullo777]

Thanks for the helpful feedback, @lmktfy and @seokho-son!

I agree that the proposal may feel unclear and too heavy at this stage.

Your suggestion helped clarify that the scope should be less about building a framework and more about giving localization teams a lightweight way to regulate and guide their use of Gen AI. That includes questions like when AI use is allowed, what should remain human-reviewed, and what practices help reduce reviewer burden and consistency issues.

A lightweight instruction.md template seems like a better starting point for that. It could help teams define their own rules for AI-assisted drafting, terminology, tone, review expectations, and safe/appropriate usage.

A PoC also makes sense. The Korean localization team may be a better starting point, since it appears to already have more established policies and practices.

I’m happy to update the issue in that direction!