Agentic engineering is a supervised learning problem.
Code = weights
Your vision = training data
Prompt = loss (approximated, you compute the real loss in your head, the prompt is your best attempt to articulate it)
Agent = weight updater
Product = model