@adamhjk I like my robots to iterate quickly the same as a human.
I've had my robots spend a lot of time improving the test speed times specifically so they can iterate faster
@zeeg Could run the evals with agent-sdk so its able to be ran with your subscription but obviously that also runs your evals within that harness which could change how things behave
@thorstenball@max_paperclips I feel like no mention of how its different from the remote sandboxing we are all used to using (e2b, daytona, etc) is a miss on the post.
@zeeg Most contracting firms you hire would charge $200+/hr and so I think it'd be reasonable to pay something similar for completed fixes if they were reasonably high quality.
@ibuildthecloud I'm using AgentSDK because my users can use their Claude subscription and it doesn't violate the ToS.
I built out an abstraction that uses pydantic-ai on bedrock for those that want API based billing and its faster, less resources, etc.
But you can't beat subscription pricing.