DataTalks.Club AI Engineering: Skill Stack, Agents, LLMOps, and How to Ship AI Products - Paul Iusztin
19 snips
Feb 6, 2026 Paul Iusztin, founding AI engineer and author of the LLM Engineer’s Handbook, outlines the full-stack AI engineer skill stack and the shift from proofs-of-concept to shipping production-grade AI. He covers when classical ML beats LLMs, building reliable agentic workflows and knowledge pipelines, using AI assistants as architects, and creating a “Second Brain” portfolio to prove end-to-end engineering value.
AI Snips
Chapters
Books
Transcript
Episode notes
LLMs Are Overused For Simple Tasks
- There's a hype push to use LLMs for everything, but classical methods still win for many tasks.
- Paul warns against over-engineering with neural nets when simpler models like XGBoost suffice.
Design Safeguards For Agents And Juniors
- Use tests, CI/CD, and environment separation to make AI codebases safe for juniors and agent assistants.
- Treat agents like interns: provide guards, monitoring, and safeguards to prevent destructive actions.
Master Knowledge Management Over Fine-Tuning
- Prioritize knowledge management, RAG, and semantic search over fine-tuning for many production systems.
- Build robust data pipelines and indexes so agents can retrieve relevant context reliably.
