Choosing Your AI Stack. Build, Buy, or Wrap.
Three options, and the wrong one is expensive in a different way each time. The decision comes down to how close the workflow sits to what makes you money.
7 min read
- Buy
- Least differentiated
- Fastest
- Wrap
- Most common fit
- Balanced
- Build
- Only for core workflows
- Slowest
Every AI project starts with the same fork.
Buy a tool, wrap a model API in your own product logic, or build the whole thing.
Teams usually pick based on engineering appetite, which is the wrong input.
The right input is how close the workflow sits to the thing customers pay you for.
Core workflows get built. Everything else gets bought or wrapped.
§01 · The problem
Where the time actually goes.
- 01Building commodity features from scratch
- 02Buying a tool for a workflow that is your edge
- 03No exit plan when a vendor changes pricing
- 04Wrapping without owning the evaluation layer
- 05Choosing based on what the team wants to build
- 06Lock-in discovered at renewal, not at signing
The question is not what is cheapest today. It is what you would be unhappy to lose control of in two years.
§02 · The mechanism
The AI workflow.
- 01
Is this workflow your differentiator?
Scoping - 02
No, and it is standard: buy it
Buy - 03
No, but your logic matters: wrap it
Wrap - 04
Yes, it is core: build it
Build - 05
Own the eval layer either way
Validation - 06
Review the decision every 12 months
Validation
A person closes the loop: the workflow ends on a human or a check against a defined outcome.
- Scoping
- A decision about what is in and out
- Buy
- Use an off-the-shelf product as it is
- Wrap
- Put your own logic around a bought model or tool
- Build
- Write it yourself: the capability is core
- Validation
- Checking the result against a defined outcome
§03 · The difference
Before vs after.
| Before | After |
|---|---|
| Decision made on engineering preference | Decision made on strategic distance |
| Core workflow outsourced to a vendor | Core workflow built and owned |
| Commodity feature built in-house for a year | Commodity features bought off the shelf |
| No evaluation layer, so quality is a guess | Evaluation owned regardless of the choice |
| Vendor pricing change forces a rebuild | Switching cost known before signing |
| Stack never revisited after the first choice | Stack reviewed annually against the same test |
§04 · The measure
Time saved.
| Task | Before | With AI |
|---|---|---|
| Time to first value | Build: months | Buy: days |
| Control over quality | Buy: low | Build: full |
| Switching cost | Buy: high | Wrap: moderate |
| Right for core workflows | Buy: no | Build: yes |
Continue reading
Tell us what you’re building.
Share a few details about what you need, and we’ll get back to you with a quotation.