Lesson 4 of 6

Build vs buy vs adapt

7 min read

You need AI in your product next quarter. Do you call an API, fine-tune a model, or train your own? Pick wrong and you either ship late or ship a toy.

Three paths to the same feature

There are really three ways to get AI capability. Buy: call a foundation model through an API — fastest, cheapest to start, least control. Adapt: take a strong base model and shape it to your job with fine-tuning or your own data — the middle ground. Build: train or host your own model — most control, most cost and expertise. Try a few needs and watch which path wins.

Buy → Adapt → Build is a ladder of rising cost and control. Most features should start at Buy; you climb only when the job truly demands it.

Default to buy, climb on purpose

The strategist's bias is to start at the cheapest rung that works. A general model behind an API now handles a huge share of jobs, so 'build our own' is rarely the right first move — it burns months and specialists before you've proven anyone wants the feature. Move to Adapt when the base model is close but needs your style, tone, or private knowledge. Move to Build only when data, latency, cost-at-scale, or control genuinely justify owning the model.

Owning a model is a big, ongoing bet. Earn the right to make it by first proving demand on a bought API — then climb only for a reason you can name.

'Adapt' often isn't fine-tuning at all — feeding the model your documents at question time (RAG) gives it private knowledge without training anything. Try the cheap version of 'adapt' first, too.