2.2. Selecting the Right AI Solution Type
💡 First Principle: Every solution type trades predictability for flexibility. Rules are perfectly predictable and brittle; predictive AI handles variety within patterns it has seen; generative AI handles open-ended language; agents act on their own across systems. The more flexible and autonomous the solution, the more value it can create and the more control it needs. The right choice is the least flexible type that solves the problem.
Why care? Because solution-type mismatch is the most expensive early mistake. Choose rules for an ambiguous problem and you get endless exceptions; choose an autonomous agent for a simple lookup and you pay for complexity and inherit risk you did not need. The exam frequently presents four options that are all "AI-ish" (a chatbot, a rules engine, a RAG assistant, an agent) and asks which fits.
A useful mental model is staffing. A checklist (rules) works for tasks anyone can follow exactly. A specialist (predictive ML) makes expert judgments within a narrow domain. A knowledgeable assistant (generative AI with retrieval) answers questions and drafts content but does not act. A delegate with authority (an agent) carries out multi-step tasks in your systems. You would not give a delegate authority to issue refunds without limits, and you would not hire one to follow a checklist.
This section also covers two responsibilities that begin the moment a solution is chosen: keeping it accurate as the world changes (monitoring for drift), and keeping the organization's use of AI tools visible and controlled (containing shadow AI).
⚠️ Common Misconception: "AI is always better than rules," "an agent is just a chatbot with a better model," "a validated model stays accurate," and "banning public AI tools stops shadow AI." Each belief picks the wrong control: too much AI, too little oversight, no monitoring, or a ban that drives usage out of sight.
