Standards across teams
Your backend, ML and frontend groups each work to their own standards, and AI agents now write code in all of them.
Where it goes wrong
Write the standard down and the part that gets agreed is formatting, because conceding formatting costs nobody anything. The questions that would cost somebody their existing design stay out of the document: which types cross between services, and what the shared library is allowed to own. What settles those is a library their code has to import in order to build.
An agent applies whichever convention sits nearest to the file it is editing, so the questions that stayed out of the document get answered again, per file, at machine speed. That is the drift you were trying to stop, now arriving faster than anyone reviews it.
How I'd handle it
I look for the smallest set of things that genuinely has to be common, and in practice it is a shared library and a typed data contract long before it is a style guide. At Line5 that meant pulling the common library into one place and replacing ad-hoc structures with typed ones, down to physical quantity types so navigation math could not silently mix units.
I put the standards where the coding agents read them, so the lint that gates a human's branch gates the agent's too.
Where I've done this
Line5's backend, ML and frontend engineers each arrived with a different idea of what good code looked like. Getting them onto shared packages, typed data and one service template is what made the standards stick, and the same rules bind the coding agents we put on the team.
Recognize this on your team?
Let's talk