Build vs buy vs wrap: the LLM stack decision

Almost every SMB asks the same question in month two: do we build our own AI, or buy a product? The honest answer is usually neither — you wrap a frontier model with your own prompts, retrieval, and evals.

Build when: the use case is core to your product, the data is proprietary and large, and you have an ML team. Buy when: a credible SaaS exists, the workflow is generic, and the integration cost is low. Wrap when: everything else — which is most of the time.