AI Is a Stack, Not a Single Technology
"AI" is too broad to be an architecture decision. A business problem may be best solved with explicit rules, statistical prediction, a neural model, generative content, a workflow or a bounded agent, and good designs often combine them. This three-part series gives non-specialists and technical leaders a problem-to-capability map, shows how layers combine safely with clear handovers, and explains why each layer needs its own evaluation. It follows Rooibos Agri Finance, a fictional agricultural finance service; its systems and figures are illustrative. Research claims were checked on 5 November 2026.
3 articles
- Part 1Agentic AI sits on a stack. It does not erase the layers below it.AI is too broad a word to make an architecture decision with. Part 1 of AI Is a Stack, Not a Single Technology maps the capabilities from explicit rules through prediction, perception and generation to workflows and agents, shows why the layers overlap rather than replace each other, and gives a problem-to-capability decision map.9 min read
- Part 2Rules, prediction, generation and agency: when the layers combineReal services rarely use one kind of AI. Part 2 of AI Is a Stack, Not a Single Technology sets out five patterns for combining rules, prediction, perception, generation and a bounded agent, the handover contract each seam needs, and the failures that happen in the joins rather than in any single layer.8 min read
- Part 3Evaluate each AI layer differentlyOne accuracy number cannot tell you whether a rule, a risk model, an image check, a generative explanation and an agent are each doing their job. Part 3 of AI Is a Stack, Not a Single Technology sets out what to measure for each layer, which metrics do not transfer, and which changes should trigger which evaluations.8 min read
Working through any of this in your own organisation?
These articles come out of delivery work rather than theory. If the problems in this series look familiar, a short scoping conversation is usually the quickest way to work out what is worth doing first.
Talk to CloudNala