Adding LLM features to an app, a hype-free take on RAG, and when an agent beats a single well-crafted prompt.
How to add a genuinely useful model-backed feature to an existing codebase while keeping it boring, testable, and cheap to remove.
A grounded look at retrieval-augmented generation for developers: when it earns its keep, how the pieces fit, and where it quietly breaks.
A concrete decision framework for choosing between a single well-crafted prompt and a multi-step agent, with the trade-offs each direction costs you.