I think there is a much-better-than-merely-non-zero chance that the rise of coding agents will also mean a rails renaissance. All of the complexity identified in the article really gets in the way of LLMs.
I think there is a much-better-than-merely-non-zero chance that the rise of coding agents will also mean a rails renaissance. All of the complexity identified in the article really gets in the way of LLMs.
It’s also the case that tried and true technologies work best for LLMs, since they have much more training on them.
Rails “convention over configuration” approach is probably also more LLM friendly. Convention can be trained into the model itself, whereas configuration takes up precious context.
I have definitely found this. A few months ago I tested Claude coding a web task in Rails vs Flask. Claude did much better with Rails - I think the opinionated Rails conventions helped the AI. However, Claude didn't complete the ask in Rails. A couple of weeks ago I revisited a similar task with one of the latest AIs. It sailed through it in Rails (didn't try Flask).
The complexity also gets in the way of junior developers but that hasn't stopped anyone yet.
So does dynamic typing to be fair.