Menu

HIA Agent Scale

Agents are defined by responsibility and relationship.

Responsibility tells you how much work the AI is trusted to carry. Relationship tells you what role it plays with the person. Together they produce a compact classification such as 4c.

See the agent map Classify a system
Put the two together

Every agent sits somewhere on this map.

Click an example name to see why it sits where it does.

The agent map
Responsibility rises vertically. Human role changes across the gradient at the top.
Answerer Assistant Thought partner Coach Companion
Level 6
Acts within delegated authorityPlans and adapts toward the goal
Level 5
Continues across timeActs from a standing assignment
Level 4
Leads a sessionDirects connected work once activated
Level 3
Completes workCarries a defined task
Level 2
Uses contextWorks from standing context
Level 1
RespondsAnswers when asked
Example

Classify a system yourself.

Choose the closest autonomy Level and relationship role. The result combines them into one classification.

Autonomy Level How much responsibility can it carry?
Relationship role What does it contribute to the human?

Need the exact definitions?

The page stays simple until you ask for more. Open either reference for the full Level or relationship definitions.

Six autonomy Levels
Level 1Chatbot

Responds to direct questions or instructions. The human initiates the exchange and decides what happens next.

Level 2Contextual Assistant

Uses standing instructions, memory, files, or a defined role, while the human still initiates each task.

Level 3Work Assistant

Helps complete a defined task, workflow, or work product after a human assigns the work.

Level 4Semi-Autonomous Agent

Once activated, can direct a complex session, coordinate steps or agents, evaluate responses, and keep the work moving toward a goal.

Level 5Supervised Autonomous Agent

Pursues an ongoing objective across time and can initiate work from a standing goal, schedule, condition, or delegated responsibility.

Level 6Fully Autonomous Agent

Interprets broad objectives, creates plans, acts, adapts, and completes work within delegated authority while humans govern and audit.

Five relationship roles
Dimension aAnswerer

Provides direct responses and information without taking on a broader role in the work or relationship.

Dimension bAssistant

Helps complete practical tasks under human direction by organizing, drafting, transforming, retrieving, or supporting the work.

Dimension cThought Partner

Improves reasoning by testing assumptions, tradeoffs, arguments, decisions, implications, and judgment.

Dimension dCoach

Helps the human improve over time through feedback, practice, adjustment, progress tracking, and attention to recurring mistakes.

Dimension eCompanion

Adds continuity, familiarity, and ongoing social or emotional presence across an extended relationship.

Where does one Level become the next?

When two Levels feel close, the difference usually comes down to who initiates the work, how much of the process the system directs, and whether responsibility continues across time.

Chatbot vs. Contextual Assistant
A Chatbot answers the question in front of it. A Contextual Assistant answers through standing instructions, memory, files, or a defined role.
Contextual Assistant vs. Work Assistant
A Contextual Assistant improves responses by applying memory, files, instructions, or role. A Work Assistant helps complete a task, workflow, or work product.
Work Assistant vs. Semi-Autonomous Agent
A Work Assistant helps complete meaningful work after being prompted. A Semi-Autonomous Agent can direct humans, coordinate other agents, evaluate responses, assign work, and manage a complex session toward a goal.
Semi-Autonomous Agent vs. Supervised Autonomous Agent
A Semi-Autonomous Agent has high autonomy inside an activated session. A Supervised Autonomous Agent has ongoing autonomy across time and can initiate work from a standing goal, schedule, condition, or delegated responsibility.
Supervised Autonomous Agent vs. Fully Autonomous Agent
A Supervised Autonomous Agent works within defined oversight, checkpoints, and escalation rules. A Fully Autonomous Agent has broader discretion to interpret goals, plan, act, adapt, and complete work within delegated authority.

Describe the system before deciding how to build or govern it.

A shared classification gives teams a clearer basis for human review, permissions, escalation rules, and operating boundaries.

See HIA agents and tools Contact HIA