Training a large language model (LLM) can cost millions of dollars, and deploying one at scale can cost millions more, but the raw model is often the wrong tool for any specific job. AI engineering bridges this gap by turning expensive research artifacts into useful products through techniques such as fine-tuning for specific domains, getting models to cite real documents instead of hallucinating, and running a billion-parameter model on a phone. A foundation model is a large model trained on broad data used as a starting point for many downstream tasks, a term coined by Stanford researchers in 2021. Training a frontier language model from scratch can require months of compute on thousands of GPUs, costing tens or hundreds of millions of dollars, while adapting an existing model might take hours on a single GPU, costing dollars. The risk is concentration, as most AI applications depend on a handful of foundation models from a handful of companies, meaning bugs, biases, or policy changes ripple through entire industries. Open-source models like Llama and Mistral provide alternatives, but today the majority of commercial AI applications still trace back to a small number of base models.