“Applied AI” is broad. It spans everything from autocomplete in an editor to fully autonomous agents that plan, execute, and report. Teams adopting AI need a way to map the landscape—to understand what they’re building and where they sit on the spectrum.

Here’s a practical taxonomy: Augmentation, Automation, and Orchestration. Each has distinct characteristics, adoption patterns, and change implications.

Augmentation

What it is: AI assists the human; the human stays in the loop for every decision. The AI suggests, drafts, or proposes; the human approves, edits, or rejects.

Examples: Code completion, AI-assisted writing, suggestion systems, copilots that offer options.

Adoption path: Lowest friction. The human remains in control. Knowledge required: how to evaluate suggestions, when to accept or reject. Desire is easier to build—it feels like “better tooling” rather than “my job is changing.”

Limitation: Throughput gains are bounded by human attention. Every output still requires a human review.

Automation

What it is: AI executes defined tasks end-to-end without human intervention. The human sets the rules and monitors; the AI does the work.

Examples: Automated testing, data pipeline processing, report generation, chatbot responses within guardrails.

Adoption path: Medium friction. The human cedes control for specific, well-defined tasks. Knowledge required: how to define and constrain the task, how to monitor for drift and failure. Desire depends on trust—people need to believe the automation is reliable and that they have recourse when it isn’t.

Limitation: Works best for tasks with clear boundaries and known inputs. Ambiguity and edge cases strain automation.

Orchestration

What it is: AI coordinates multiple steps, tools, and sometimes other agents. The human sets goals and intervenes at key points; the AI plans, delegates, and executes.

Examples: Agentic development (plan → code → test → iterate), multi-step research assistants, autonomous workflows that hand off to humans when stuck.

Adoption path: Highest friction—and highest leverage. The human shifts from executor to strategist and validator. Knowledge required: prompt design, evaluation, handoff protocols, failure recovery. Desire is harder: roles change more dramatically. But the upside is largest: less queue, more craft.

Limitation: Complexity. Orchestration systems have more failure modes, more coordination overhead, and require stronger mental models to operate effectively.

Using the Taxonomy

Teams can map their initiatives: “We’re starting with Augmentation in the editor, moving to Automation for tests, and piloting Orchestration for feature development.” That mapping clarifies what knowledge and ability each phase requires—and where ADKAR needs to do the most work. It also sets expectations: Orchestration isn’t “Augmentation with more steps.” It’s a different way of working, and it deserves different preparation.