Artificial Intelligence (AI)

Artificial intelligence is the field concerned with building systems that perform tasks we associate with human intelligence — recognising images, understanding and producing language, planning, deciding, and learning from experience. It is an umbrella, not a single technique: expert systems that encode hand-written rules, the statistical machine-learning methods that dominated the 2000s, and the deep neural networks behind today's language and image models are all AI. What changed in the 2020s is not the definition but the capability curve, as transformer models trained on internet-scale data crossed the threshold from research demo to something you can put in front of a paying customer. For a SaaS builder the practical meaning is narrower than the headlines suggest. Almost nothing shipped commercially is a general reasoning system; it is a specific model doing a specific bounded job — classifying support tickets, drafting a product description, extracting fields from an invoice, ranking search results. The engineering discipline is therefore mostly about scoping: choosing a task where a probabilistic output is acceptable, defining what "good enough" means numerically, and designing the surrounding product so a wrong answer is recoverable rather than catastrophic. That last point separates features that survive contact with users from demos that don't. A summariser that occasionally invents a detail is fine when a human reads the summary next to the source; the same model auto-approving refunds is not. So the useful questions when someone says "we should add AI" are: what decision are we automating, what does being wrong cost, and who catches it. The vocabulary matters too, because vendors use it loosely. "AI" on a pricing page may mean a frontier language model, a small classifier, or a lookup table with good marketing. Ask which, and ask what happens when it fails — the answer tells you far more about the product than the label does.