In practice, for many useful applications, many of the various obvious problems with AI agents (drift, hallucination, compounding errors) are more solvable than they are in theory Clever prompting, tool use, constrained topics,
LLM judges & organizational process close some gaps

→ View original post on X — @emollick