In Haskell, I see mtl and algebraic effects (say freer-simple) as giving you the same kind of expressiveness. The difference to me is that for mtl you need to figure out and abstract a new type class for every kind of effect and then write n^2 instances. While the freer monad construction needs only a single data type (often a GADT) and some glue function (calling send on constructors of said data type), and you are off to the races.
The algebraic reason for this is that effects are combined with sum, which is commutative up to isomorphism. While transformers are not naturally commutative, so mtl must write all the commuters as instances.
This, along with the reinterpret functions means that you can quickly spin up custom effects for your program, which do exactly what you need to express your program logic. Then all the glue coddle to make your program interact with the real world becomes a series of handlers, usually refining in several steps until you reach IO.
When I have used mtl, I end up only using the standard monad classes, and then I have to remember the semantics of each one in my domain.
The algebraic reason for this is that effects are combined with sum, which is commutative up to isomorphism. While transformers are not naturally commutative, so mtl must write all the commuters as instances.
This, along with the reinterpret functions means that you can quickly spin up custom effects for your program, which do exactly what you need to express your program logic. Then all the glue coddle to make your program interact with the real world becomes a series of handlers, usually refining in several steps until you reach IO.
When I have used mtl, I end up only using the standard monad classes, and then I have to remember the semantics of each one in my domain.