The Cloudcast Will there be a market for expert AI agents?
48 snips
Dec 14, 2025 Delve into the intriguing world of AI expert agents, exploring the challenges of training them in specialized domains like accounting and law. Discover the nuances of teaching these agents through case studies and handling ethical dilemmas. The discussion navigates who will benefit from these advancements and how such agents might compare in cost to human experts. Plus, hear about the potential for agents to build memory and experience while protecting client privacy, and ponder which domains may see the first full automation.
AI Snips
Chapters
Transcript
Episode notes
Start With Academic Foundations
- Aaron Delp argues expert AI agents would start from frontier language models then be trained on domain coursework and textbooks.
- He views formal education + textbooks as the first essential layer for domain expertise.
Train With Real Case Studies
- Delp says well-defined case studies are needed to teach context and application beyond theory.
- Case studies reveal corner cases and tradeoffs that pure coursework misses.
Corner Cases Drive Most Work
- He highlights corner cases and error handling as the bulk of training effort and likens it to programming robustness.
- Experience-driven failure modes are what make agents behave like seasoned professionals.