One of the easiest mistakes to make in enterprise AI is to assume that the same architecture should work everywhere.
A hospital, a software company, an automotive manufacturer and a professional services firm may all use AI for writing, search, summarization, analysis and recommendation. But the information being handled, the consequences of being wrong, the amount of cognition required and the rules governing how information can move can be completely different.
Industry is the first filter. Workflow is the final one.
BOUNDARIES
Start with information movement
What information is allowed to cross which boundary, under what controls, for what purpose?
KNOWLEDGE
The organization provides governed truth
A general model can probably produce a sensible policy. That does not mean it knows the company’s policy. The organization’s approved policy must remain the source of truth.
The model provides general cognitive capability. The organization provides governed truth.
ECONOMICS
Use only as much intelligence as the task requires
The strongest model is not automatically the right architecture.
A senior engineer asking AI to analyse one difficult failure every fortnight has a different cost structure from a production workflow classifying 100,000 records.
ARCHITECTURE
Design from business conditions outward
Business objective → information sensitivity → compliance and control → organizational knowledge → cognitive complexity → volume → consequence → architecture.
The architecture should follow the business.