AI Agents Have an Information Cocoon Problem No One Talks About
blog.trine.devIn my experience, the more verification and validation the human does for new projects has allowed me to surpass the ceiling because I feel I have provided the needed intent and directives to allow the agent to build workflows around the intent and get to highly efficient and effective.
The experiments and pushing the bounds is important, but the extent that the agents can be made productiv, while left to their own devices, is dubious at this point. Human in the loop is the way, combine the best of both.