In a new retrospective, Eugene Yan shares what a year of building with large language models has taught him. The piece is organized around three levels of concern: the tactical nuts and bolts of implementation, the operational day-to-day of running systems, and the longer-term business strategy that shapes why and how LLMs are used.
Because this is a single source, there are no competing views to compare. The article's main thrust is that successful LLM work is not just about model quality or prompt design; it also depends on operational discipline and strategic alignment. The lessons are drawn from real building experience rather than abstract theory.
The author's framing suggests that teams should pay attention to all three layers at once. Tactical wins can be undermined by operational gaps, and operational stability matters little if the work is not tied to a coherent business strategy. The retrospective serves as a practical field guide for others navigating similar challenges.