Great read! How much do you think chiseling AI slop from a big agent run could be done with a series of iterations or more specialized agents? Always curious what people feel the limits of current AI are and what distinguishes the human touch