Porting got cheap
An AI agent could move one of your services to another language this quarter, and you're wondering if you should.
Where it goes wrong
Generating the new code is the part that got cheap. What stays expensive is everything the old language was quietly supplying: libraries that exist and are stable, a profiler your people can already read, a release path, and someone who can debug the result at three in the morning in a language they are still learning.
A Rust component I wrote at Redis is a fair example. The translation was the small part. The cost sat in the ecosystem underneath it moving while the code above it had to keep running, and a coding agent shortens none of that.
How I'd handle it
I price the parts an agent does not touch. What does the target language's ecosystem hand you today, and what will it leave you to build yourself? Plus the unglamorous half: how the artifact builds and ships, and who is on call for it. That price is usually the whole decision, and almost nobody argues about it.
I also want a boundary where one component can go first and answer those questions in production at small scale. A working precedent settles the larger argument far more cheaply than a proposal does.
Where I've done this
My Rust bet at Redis went in one component at a time inside a mature C and Python product, and an early one was a piece of the cluster software that had to keep working. That precedent carried further than the component did: when the company later weighed which language to rebuild its cluster management in, I was asked, and that daemon was most of what I argued from.
Recognize this on your team?
Let's talk