build vs buy vs llm, visualized
Build it, buy it, or ask a model to do it?
For decades this was a two-way call between writing the software yourself and paying a vendor. Now there's a third option: describe the job and let an LLM do it at runtime. AI also changed the cost of the first two. These pages lay out the pros and cons of each path and show how the weights moved.
Three paths to the same capability
Build
Your team writes, ships and owns the code.
+ strongest
Perfect fit
Differentiation & moat
− weakest
Slow to first value
High upfront cost
Buy
Rent a product someone else builds and runs.
+ strongest
Fastest to production
Shared R&D cost
− weakest
Lock-in & switching cost
Pricing scales against you
LLM
Describe the job; let a model do it at runtime.
+ strongest
Prototype in hours
Handles the fuzzy long tail
− weakest
Non-deterministic
Per-call cost & latency
01→
Build
Own the code
02→
Buy
Rent the product
03→
LLM
Let a model do the job
04→
Cost over time
TCO curves & break-even
05→
Decision matrix
Weight what matters to you
06→
What AI changes
How the weights moved
07→
It's a stack
Buy, build & LLM together
08→
Decide
Seven questions → a call
- the short versionBuilding got cheap. Coding agents cut the cost of writing and maintaining code, so the Build side weighs less than it used to.
- the short versionCommodity SaaS got weaker. A thin CRUD tool can be rebuilt in days. Vendors whose moat is trust, data or a network are stronger than before.
- the short versionLLMs became an option of their own. Messy, open-ended work that used to need people or brittle rules can be handed to a model, as long as you can check the output.