Agents are perfectly capable of learning. Why would the model need to learn? The harness and tooling are all that matter.
They aren't, it's just populating context window until "memory" is pushed out.
They aren't, it's just populating context window until "memory" is pushed out.