🎬 AI Agent Specialization. RAG vs Fine-tuning — T3chFest 2026
Video Details
Channel: T3chFest URL: https://www.youtube.com/watch?v=weNLIRJutC8 Relevance: ⭐⭐⭐⭐
Summary
Conference presentation at T3chFest 2026 comparing RAG-based specialisation vs fine-tuning for domain-specific agents. Key finding: for tasks with evolving data (like Jira issue classification), RAG consistently outperforms fine-tuning because fine-tuned models become stale. For stable domains, fine-tuning is superior.
PUMA Relevance
Directly supports PUMA’s decision to use RAG (Stage 4) rather than fine-tuning for the triage agent. The ‘evolving data → RAG superior’ finding applies exactly to PUMA: Jira issue vocabularies and priorities evolve with team practices, making a RAG approach over the historical issue corpus more robust than a fine-tuned classifier.