Types of AI agents
Six distinct rungs, each with more capability, cost, variance, and risk than the last — from a bare model to a fully governed multi-agent system. Most confusion about "AI agents" is a failure to distinguish these.
Rung 1
AI Model
Can: Generate text, classify, reason within a single inference
Can't: Remember, act, verify, or change anything in the world
Rung 2
AI Assistant
Can: Help a person think, draft, summarise, answer
Can't: Work unattended; complete a process
Rung 3
AI Automation
Can: Execute known steps reliably and cheaply at scale
Can't: Handle situations its designer did not anticipate
Rung 4
AI Agent
Can: Pursue a goal across many steps in a changing environment
Can't: Exceed its permissions, knowledge, tools, or governance
Rung 5
Multi-Agent System
Can: Parallelise, specialise, delegate, review each other
Can't: Escape the coordination cost it creates
Rung 6
Holistic Agent System
Can: Own an outcome, be measured, audited, and improved
Can't: Absorb human accountability
Only move up a rung when the problem genuinely requires it — capability and risk rise together. This is taught in full in Module 1 of the free course.