Intelligence is an AI system’s capability to infer, generate, decide, or perform tasks requiring adaptive reasoning. In product work, it is one component of a larger system alongside data and user experience. section 3.0
Choose an intelligence approach from the learning problem rather than from current fashion. Rule-based, symbolic systems suit explicitly specified behavior; machine-learning, or neural, approaches suit problems whose patterns must be learned. Generative AI is only one possibility. section 3.2 section 3.0
Match the system type to the work it must perform:
- Predictive or analytical AI analyzes data to make predictions for clearly defined problems, often using labeled data. section 3.4
- Generative AI creates new content by learning patterns from training data. section 3.4
- Agentic AI automates multistep workflows and can combine generative and predictive models with tools. section 3.4
A practical model choice
A team initially considers GPT-4o for a simple analytical task, then recognizes that logistic regression can provide a smaller, faster, fairly accurate solution on its own infrastructure, reducing privacy concerns. section 3.0
Good intelligence choices balance capability with privacy, resources, user needs, and delivery risk. Treat outputs as probabilistic: communicate limitations, prototype and test with realistic scenarios, add safeguards, and plan evaluation and iteration after launch. section 1.5 section 3.0
