RuboCop's place in an AI/agentic future

[andyw8]

👋 I'd like to start a discussion about how RuboCop fits into a future where most code is written by AI and not humans. Some initial thoughts:

• RuboCop can provide a very effective set of guardrails, helping to ensure Ruby written by AI is idiomatic.
• I've seen AI pick up idioms from other languages (example), so there may be new cops needed to protect against that.
• Question: Should linting messages give extra direction to agents, e.g. to suggest refactoring if there is a metrics cop violation? Such messages may be overwhelming for humans, but valuable for AI.

[bbatsov]

I've seen AI pick up idioms from other languages (rubocop/ruby-style-guide#977), so there may be new cops needed to protect against that.

Very good point!

Question: Should linting messages give extra direction to agents, e.g. to suggest refactoring if there is a metrics cop violation? Such messages may be overwhelming for humans, but valuable for AI.

I think we can have some more detailed formatters for agents, plus @koic has just started working on #14911, which will also help agentic flows in Ruby.

But I totally agree that we need to spend some time thinking how to make the best of RuboCop in the agentic programming world. Seems that the approach of most people to programming in the past couple of years changed more than in the past decade.

[andyw8]

I see you added some agent rules in #14912, that should be a good help too. 👍

[corsonknowles]

I've been thinking about this for a while (not internal to Rubocop, but downstream as a user):

• https://engineering.gusto.com/our-super-power-is-our-community-and-our-standards-092ffefe26ac

Since writing that, we published:

• https://github.com/gusto/rubocop-gusto

There's no specifically agentic rules there yet, but I will look around to see if we have anything more we can add.