Lesson 3 of 6
Build vs buy vs fine-tune
7 min read
You need an AI feature. Do you call a ready-made model, teach one your style, or train your own? Pick wrong and you burn months you didn't have.
Three ways up the same hill
There are three ways to get an AI capability. Buy: call a general foundation model as-is — fastest, cheapest, least control. Fine-tune: take a ready model and teach it your style and data — more effort, more fit. Build: train or heavily customize your own — most control, most cost. Pick a need below and watch which one wins.
Buy, fine-tune, or build: the same capability has three price tags. The right one depends on the job, not on which sounds most impressive.
Start at Buy, climb only when forced
The rule of thumb: start at the cheapest rung that works. A general model handles a surprising amount as-is. Reach for [fine-tuning](glossary://finetuning) only when the base is close but misses your style or domain; reach for building your own only when data, latency, or control genuinely demand it. Most teams over-climb — they build when they could have bought.
Prompting and retrieval come first. Often what looks like 'we need to fine-tune' is really 'we haven't written a clear prompt or given it the right context yet'.
Every rung up the ladder costs time and money. It should be forced by the job — never chosen for the thrill of building.
The shape of it
- —Buy — call a general model as-is. Fast, cheap, least control. Start here.
- —Fine-tune — teach a ready model your style and data when the base is close but off.
- —Build — train or deeply customize your own only when data, latency, or control demand it.
- —Pick the cheapest rung that does the job — most teams climb too high.
A general model already answers your generic FAQ well. What's the right move?
Continue in the app
Take the whole Becoming an AI Engineer course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.