Generative AI → AI Agent → Agentic AI
Overview
These aren’t separate categories with rigid boundaries, they represent successive stages in the evolution of AI capabilities. Understanding the progression clarifies what each can do and what governance it requires.
Generative AI
What it is: Systems (LLMs, image models) that generate novel outputs in response to user prompts.
Behavior: Reactive. Produces content when prompted. Does not independently pursue objectives or interact with external environments unless explicitly directed.
Examples: ChatGPT responding to a question, DALL-E generating an image from a prompt.
Governance implication: Human initiates every action. Quality control is on the output.
AI Agent
What it is: Generative AI extended with tool use, function calling, sequential reasoning, and limited autonomy.
Behavior: Can perform multi-step, goal-directed tasks. Retrieves real-time information, interacts with APIs and software systems. Executes well-defined functions within bounded scope. Autonomy limited to single workflow or session.
Examples: A coding assistant that can run tests and iterate on code. A research agent that searches, synthesizes, and formats findings.
Governance implication: Human sets the goal and boundaries. Agent executes within constraints. Human reviews output.
Agentic AI
What it is: Multiple specialized AI agents collaborating, communicating, planning dynamically, and allocating sub-tasks to achieve complex shared objectives.
Behavior: Persistent memory across sessions. Dynamic task decomposition. Multi-agent orchestration. Emergent behaviors in unstructured environments. High adaptability and coordinated autonomy.
Examples: A system where one agent identifies market opportunities, another generates product concepts, a third runs experiments, and they coordinate toward a product launch objective.
Governance implication: Humans set objectives and constraints, but agents determine execution paths. Oversight shifts from reviewing outputs to monitoring behaviors and intervening on exceptions.
Key Differences
| Dimension | Generative AI | AI Agent | Agentic AI |
|---|---|---|---|
| Initiative | Reactive | Goal-directed within session | Self-directed across sessions |
| Scope | Single prompt-response | Single workflow | Complex multi-stage objectives |
| Memory | None/limited | Session-based | Persistent |
| Coordination | None | None | Multi-agent collaboration |
| Autonomy | None | Bounded | Coordinated/emergent |
| Governance | Output review | Constraint setting | Behavior monitoring |
When Each Applies
- Generative AI: Content creation, brainstorming, translation, summarization
- AI Agent: Repetitive multi-step workflows, research synthesis, code completion
- Agentic AI: Complex problem-solving requiring adaptation, experimentation, and coordination across domains
The Progression Insight
Most organizational AI maturity journeys start with generative AI, graduate to AI agents for workflow automation, and eventually encounter the governance challenges of agentic AI. The shift from agent to agentic is where traditional PM frameworks break down.
Related: 05-atom—agentic-ai-definition, 07-molecule—pm-as-ecosystem-orchestrator, 05-atom—automation-augmentation-paradox