Haskell works pretty well with agents, particularly when the agent is LSP-capable and you set up haskell-language-server. Even less capable models do well with this combo. Without LSP works fine but the fast feedback loop after each edit really accelerates agents while the intent is still fresh in context