Arch Forum 2025-06-12

Participants: Backend devs, Andy & Victor

Agenda

• KYC AI architecture (Alex)
• AI usage guidelines

Summary

KYC AI architecture

Alex had a walkthrough of how the new AI-powered Address Verification works in KYC.

Presentation here: https://www.figma.com/board/4wifkJIy3CfSox95fWpOFk/KYC-Gemini

AI usage guidelines

We also discussed two ideas around how to use AI, inspired by this blog https://diwank.space/field-notes-from-shipping-real-code-with-claude

Mark in PRs how AI helped with the code

On a high level it seems like a good thing, at least for statistics. There was no strong belief that it would be helpful to know when reviewing code (even if it wouldn't hurt).

However, the difficult part is how it could be implemented. Several people raised their concerns about how to judge what the AI did, for example when you "discuss" solutions with AI, or when you let it generate code that you later partially refactor/rewrite.

Do not let AI write tests.

We are not sure about this. At the moment it feels like our AI usage is not big enough to motivate this, and ultimately the developer must be responsible for all code they commit regardless of whether it's AI generated or human written. But it is an interesting idea!