I would like to see 2 AI native companies try this - even giving the agent(s) some sort of read/write permissions within a trusted env.