Six articles on choosing the kind of AI system a job calls for, sorted not by the name on the slide but by two questions: who decides, and who acts. Each part takes one family, places three generic jobs on it, names what the platform from the agent stack blueprint has to provide, and quotes the article of law that applies. Read in order; the map in part 1 is the reference for all the others.
| Part | Article | Reading | Link |
|---|---|---|---|
| Part 1 | The map: who decides, who acts, and how far the system goes on its ownEight families sorted by two questions, a five-level autonomy scale, the worst action, the data's age, and the platform layer that carries the risk. Three jobs placed on the map, with the article of law for each placement. | 14 min | Open |
| Part 2 | AI without a language model: prediction, recommendation, perception, optimisationWhat already runs everywhere, what it asks of the platform, and why it is not "less" than the rest. | 15 min | Open |
| Part 3 | Generating and assisting: generative AI, integrated copilots, small specialised modelsThe person reads, the risk is the content. | 17 min | Open |
| Part 4 | Answering on your own documents: RAG, agentic RAG, graph RAGThe risk becomes access to sources, and freshness. | 16 min | Open |
| Part 5 | Acting within a perimeter: the AI agent with its tools, and the augmented workflow as the migration pathThe risk is the action. The perimeter is outside the model, and the read / write / commit label is yours to keep. | 20 min | Open |
| Part 6 | Pursuing a goal with several agents: agentic AI, delegation, intentThe unit of governance is the handover; the most demanding regime, and the one sold first. | 20 min | Open |