Core Disciplines

How engineers reason across the handbook.

The chapters teach in sequence. The disciplines are the lenses used across that sequence when reviewing AI system decisions.

Discipline

Prompt Engineering

Turns system intent into maintainable instruction interfaces.

Use this lens when a behavior change appears to be a wording change but actually changes product responsibility.

Discipline

Controls what information the model receives for a task.

Use this lens when correctness depends on what the model knows, what it should not know, or what source it should trust.

Discipline

Wraps model capability in deterministic runtime control.

Use this lens when the system must decide whether to accept, retry, route, escalate, or stop.

Discipline

Converts expected behavior into evidence for shipping and changing the system.

Use this lens when a team says the system works but cannot show evidence matched to risk.

Discipline

Designs responsibility boundaries across models, tools, humans, and workflows.

Use this lens when adding autonomy, connecting tools, or choosing between RAG, copilot, agent, and workflow patterns.

How to use them

Which discipline owns the failed behavior?
Which discipline is being overloaded because another one is missing?
Which discipline supplies the evidence that the change is safe?
Which discipline must review the next increase in autonomy, exposure, or blast radius?