Early AI developers—especially those working on modern deep learning and Large Language Models (LLMs)—were deeply surprised by what their models could do. In fact, "surprise" is practically a core theme in the history of modern artificial intelligence.

The root of this surprise comes down to a fundamental shift in how software is made: instead of writing explicit, line-by-line instructions, engineers built architecture that learned from data. 

Developers found themselves caught off guard by their own creations.