---

name: ai-collaboration-modes description: "Select appropriate human-AI collaboration modes: Automation (AI acts independently), Augmentation (AI assists human work), or Agency (AI takes delegated responsibility)." version: "1.0.0"

Overview

AI Collaboration Modes defines three distinct patterns for human-AI work: Automation, Augmentation, and Agency. Each mode has different characteristics, appropriate uses, and governance requirements.

Core Principle: The mode of collaboration should match the task characteristics, not default to a single pattern.

---

When to Use This Skill

• Designing AI-assisted workflows
• Choosing how to apply AI to a task
• Setting appropriate oversight levels
• Defining accountability structures
• Scaling AI across an organization

---

The Three Modes

Mode 1: Automation

Definition: AI performs tasks independently with minimal human involvement.

Pattern:

CopyInput → [AI Process] → Output → [Use]
           ↓
    (Monitoring only)

Characteristics:

• High volume, low variance tasks
• Standardized inputs and outputs
• Clear success criteria
• Low cost of individual errors
• Aggregate quality matters more than individual quality

Examples:

• Email categorization
• Data format conversion
• Basic content tagging
• Routine document generation
• Simple customer query responses

Appropriate when:

• Task is repetitive and well-defined
• Output can be monitored at aggregate level
• Individual errors are recoverable
• Human review would be inefficient
• Quality requirements are well-specified

Governance requirements:

• Automated quality monitoring
• Exception handling for edge cases
• Regular sampling and audit
• Clear escalation criteria
• Performance dashboards

Risk profile:

• Low risk per task
• Aggregate risk if systematic errors
• Risk of drift without monitoring
• Complacency risk over time

---

Mode 2: Augmentation

Definition: AI assists human work, with humans retaining control and decision-making.

Pattern:

CopyInput → [AI Draft/Analysis] → [Human Review/Edit] → Output
                ↓                       ↓
         (Assists human)        (Human decides)

Characteristics:

• Human expertise is essential
• AI accelerates or enhances human work
• Judgment required for final output
• Quality depends on human oversight
• Human retains accountability

Examples:

• Document drafting with human editing
• Research synthesis for human analysis
• Code suggestions for human implementation
• Data analysis for human interpretation
• Decision support (not decision-making)

Appropriate when:

• Human judgment is needed
• Stakes are significant
• Context matters
• Output requires accountability
• Quality is critical

Governance requirements:

• Clear human review process
• Documented approval workflow
• Training on AI limitations
• Verification protocols
• Accountability assignment

Risk profile:

• Dependent on human oversight quality
• Risk of automation complacency
• Risk of output authority bias
• Mitigated by human in loop

---

Mode 3: Agency

Definition: AI takes delegated responsibility for achieving objectives, with human oversight at strategic level.

Pattern:

CopyObjective → [AI Planning] → [AI Execution] → [AI Evaluation] → Result
                ↓                                    ↓
        (Human approves plan)              (Human reviews result)

Characteristics:

• Multi-step, goal-oriented tasks
• AI determines approach within constraints
• Human provides objectives and boundaries
• Outcomes evaluated against goals
• Human oversight is strategic, not tactical

Examples:

• Research and analysis projects
• Complex document creation
• Multi-step workflow execution
• Investigation and synthesis
• Project-scoped tasks

Appropriate when:

• Task is goal-oriented, not script-defined
• Multiple approaches are valid
• Human time is the constraint
• Results can be verified
• Stakes warrant careful oversight

Governance requirements:

• Clear objective specification
• Defined boundaries and constraints
• Approval gates for significant actions
• Result verification process
• Clear accountability structure

Risk profile:

• Higher individual task risk
• Dependent on objective clarity
• Risk of scope creep
• Requires strong verification

---

Mode Selection Framework

Selection Criteria

Decision Tree

CopySTART
  │
  ▼
Is the task repetitive with clear rules?
  │
  ├─ YES → Is individual error cost low?
  │         │
  │         ├─ YES → AUTOMATION
  │         │
  │         └─ NO → AUGMENTATION
  │
  └─ NO → Does task require human judgment?
           │
           ├─ YES → AUGMENTATION
           │
           └─ NO → Is task goal-oriented with clear success criteria?
                    │
                    ├─ YES → AGENCY
                    │
                    └─ NO → AUGMENTATION (default to human control)

Mode Selection Template

CopyMODE SELECTION ANALYSIS

Task: [Description]

CHARACTERISTICS:
- Complexity: [Low/Medium/High]
- Volume: [Low/Medium/High]
- Judgment needed: [Minimal/Moderate/Significant]
- Error cost: [Low/Medium/High]
- Standardization: [Low/Medium/High]

CANDIDATE MODES:
□ Automation - because: [rationale]
□ Augmentation - because: [rationale]
□ Agency - because: [rationale]

SELECTED MODE: _______________

GOVERNANCE FOR THIS MODE:
- Oversight level: [Describe]
- Review process: [Describe]
- Quality checks: [Describe]
- Escalation criteria: [Describe]

---

Mode Transitions

When to Shift Modes

Automation → Augmentation:

• Error rates increase
• Edge cases become common
• Stakes increase
• Regulations change

Augmentation → Automation:

• Process becomes stable
• Human review adds little value
• Volume increases
• Confidence in AI quality

Augmentation → Agency:

• Human time is the bottleneck
• AI demonstrates reliability
• Goals can be clearly specified
• Results can be verified

Agency → Augmentation:

• Results are inconsistent
• Objectives are unclear
• Stakes increase
• More control needed

---

Hybrid Patterns

Multi-Mode Workflow

CopyEXAMPLE: Content Creation Pipeline

Stage 1 (Automation):
  - Topic categorization
  - Initial research compilation
  - Basic fact extraction

Stage 2 (Agency):
  - Draft creation to specification
  - Structure and organization
  - Reference integration

Stage 3 (Augmentation):
  - Human review and editing
  - Tone and voice adjustment
  - Final approval

Stage 4 (Automation):
  - Formatting
  - Publishing
  - Distribution

Mode by Risk Level

CopyRISK-BASED MODE ASSIGNMENT

Low risk tasks → Automation
- Internal documentation
- Routine communications
- Data formatting

Medium risk tasks → Augmentation
- External communications
- Analysis and recommendations
- Decision support

High risk tasks → Augmentation with enhanced review
- Legal/compliance content
- Financial reporting
- Customer-facing commitments

Variable risk tasks → Agency with gates
- Research projects
- Complex analysis
- Multi-step workflows

---

Practices

Mode Audit

For existing AI use:

CopyMODE AUDIT

Current AI Applications:
| Task | Current Mode | Appropriate? | Recommended |
|------|--------------|--------------|-------------|
| [Task] | [Mode] | [Yes/No] | [Mode] |

Gaps identified:
1. [Gap]
2. [Gap]

Changes needed:
1. [Change]
2. [Change]

Mode Governance Matrix

CopyGOVERNANCE BY MODE

| Element | Automation | Augmentation | Agency |
|---------|------------|--------------|--------|
| **Oversight** | Aggregate | Per-output | Per-objective |
| **Review** | Sampling | All outputs | Results |
| **Approval** | Exception-based | Before use | Before execution |
| **Accountability** | System owner | Human reviewer | Delegator |
| **Monitoring** | Dashboards | Review logs | Outcome tracking |

---

Assessment Criteria

Mode Selection Proficiency When:

• Can identify appropriate mode for any task
• Understands governance requirements per mode
• Can design hybrid multi-mode workflows
• Recognizes when mode transitions are needed
• Has documented mode assignments for key workflows

---

Common Mode Failures

Failure 1: Automation of High-Stakes Tasks

Wrong: Automating decisions that require judgment Right: Use augmentation with human review for significant decisions

Failure 2: Augmentation Overhead

Wrong: Human review of every trivial AI output Right: Automate routine tasks, focus human time on value

Failure 3: Agency Without Boundaries

Wrong: "Just figure it out" without constraints Right: Clear objectives, defined boundaries, approval gates

Failure 4: Static Mode Assignment

Wrong: Same mode forever regardless of results Right: Regular review and mode adjustment based on performance

---

Related Skills

• ai-workflow-integration — Implementing modes in workflows
• ai-system-governance — Governance by mode
• ai-cognitive-readiness — Knowing when AI should act independently

---

Learn More

• Mode Selection Examples
• Governance Templates by Mode