The Blended Workforce Model¶

Overview¶

The Blended Workforce Model is ZORBA's answer to the question that no traditional framework addresses: how do humans and AI agents co-exist as a unified workforce within an enterprise?

This document defines:

Agent types — the roles agents play in enterprise operations
Collaboration patterns — how humans and agents work together
Trust and autonomy levels — the graduated spectrum of agent independence
Workforce composition design — how to make deliberate decisions about the human/agent mix

This is not theory. This is a practical model for designing, governing, and evolving a workforce where some of the participants are not human.

Part 1: Agent Types¶

Not all agents are alike. ZORBA defines four fundamental agent types based on their relationship to human workers and their scope of autonomy.

1.1 Autonomous Agents¶

Definition: Agents that execute defined work independently, within bounded authority, without requiring human involvement for each instance.

Characteristics: - Operate within clearly defined parameters and constraints - Handle routine decisions within their authority boundary - Self-monitor performance and quality - Escalate only when encountering situations outside their defined scope - Produce audit-grade execution logs

Enterprise examples: - Invoice processing agent that validates, matches, and approves payments within defined thresholds - Monitoring agent that detects anomalies and triggers response protocols - Content moderation agent that applies policy to user-generated content - Data quality agent that continuously validates and cleanses data pipelines

ZORBA layer affinity: Primarily L5 (Activities) and L6 (Work)

Governance requirement: Defined authority boundaries, continuous performance monitoring, human-reviewable decision logs, automated escalation triggers.

1.2 Co-Pilot Agents¶

Definition: Agents that work alongside a specific human, augmenting their capabilities, preparing materials, and handling supporting tasks while the human retains decision authority.

Characteristics: - Paired with a human worker (1:1 or 1:many) - Understand the human's context, preferences, and patterns - Prepare options and recommendations rather than making final decisions - Handle preparatory and follow-up work around human decision points - Adapt their behaviour based on human feedback

Enterprise examples: - Executive co-pilot that prepares meeting briefs, drafts communications, and tracks action items - Sales co-pilot that researches prospects, drafts proposals, and updates CRM - Legal co-pilot that reviews contracts, flags risks, and suggests clause alternatives - Engineering co-pilot that writes code, reviews PRs, and maintains documentation

ZORBA layer affinity: L2 (Objectives) through L5 (Activities) — wherever the paired human operates

Governance requirement: Clear scope definition, human override at all times, transparency about agent actions taken on behalf of the human, data access controls.

1.3 Specialist Agents¶

Definition: Agents with deep expertise in a narrow domain, called upon for specific tasks that require specialist knowledge or capability.

Characteristics: - Deep capability in a bounded domain - Invoked by humans or other agents for specific tasks - Stateless between invocations (or with managed state) - Provide expert output but do not own end-to-end processes - May be shared across teams and domains

Enterprise examples: - Financial modelling agent that builds and stress-tests financial projections - Translation agent that localises content across markets - Compliance checking agent that validates activities against regulatory requirements - Data analysis agent that performs statistical analysis on demand

ZORBA layer affinity: L5 (Activities) — called into processes as needed

Governance requirement: Input/output validation, capability certification, version control, access controls for sensitive specialist functions.

1.4 Orchestrator Agents¶

Definition: Agents that coordinate and manage workflows involving multiple humans and/or agents, ensuring work flows correctly through defined processes.

Characteristics: - Manage the flow of work, not the content of work - Route tasks to appropriate performers (human or agent) - Monitor deadlines, SLAs, and dependencies - Handle handoffs between participants - Maintain process state and provide visibility

Enterprise examples: - Workflow orchestrator managing an approval chain across departments - Incident response orchestrator coordinating between detection, analysis, and resolution agents/humans - Onboarding orchestrator managing the sequence of activities for new employee setup - Campaign orchestrator coordinating marketing activities across channels

ZORBA layer affinity: L4 (Processes) — managing the flow at the process level

Governance requirement: Routing logic transparency, SLA enforcement rules, escalation authority, visibility to all process participants, no ability to modify process definitions without human approval.

Agent Type Summary¶

Type	Autonomy	Scope	Human Relationship	Primary Layer
Autonomous	High	Bounded	Independent within authority	L5–L6
Co-Pilot	Medium	Adaptive	Paired with human	L2–L5
Specialist	Variable	Narrow & deep	On-demand invocation	L5
Orchestrator	Medium-High	Cross-functional	Coordinates humans & agents	L4

Part 2: Collaboration Patterns¶

Humans and agents don't just "work together" — they collaborate in specific, definable patterns. ZORBA identifies six fundamental collaboration patterns.

2.1 Delegation¶

Pattern: A human assigns a task or set of tasks to an agent, defining the expected outcome, constraints, and reporting requirements.

Characteristics: - Clear task definition from human to agent - Defined success criteria and constraints - Agent has autonomy within the delegated scope - Human retains accountability for the outcome - Agent reports back on completion or exception

When to use: Routine tasks with well-defined parameters; tasks where agent execution is more efficient; work that needs to happen at scale or speed beyond human capacity.

Governance consideration: The delegating human must have authority to delegate the task, and the agent must operate within the authority it has been granted, not the authority of the delegating human.

2.2 Supervision¶

Pattern: An agent performs work while a human monitors, reviews, and intervenes as needed. The human does not perform the work but validates and corrects agent output.

Characteristics: - Agent is the primary performer - Human reviews output (all, sampled, or exception-based) - Human can intervene, correct, or override at any point - Agent learning may be informed by human corrections - Supervision intensity may decrease as trust increases

When to use: Early-stage agent deployment; high-stakes activities; regulated processes requiring human oversight; trust calibration periods.

Governance consideration: Supervision ratios must be defined (e.g., 100% review, 10% sampling, exception-only). Supervision logs must capture what was reviewed, by whom, and what actions were taken.

2.3 Peer Collaboration¶

Pattern: Humans and agents work together as peers on a shared task, each contributing their distinct capabilities.

Characteristics: - Shared objective with distinct contributions - Human provides judgement, creativity, relationships, context - Agent provides speed, data processing, consistency, breadth - Iterative exchange — each builds on the other's output - Neither is subordinate; both are essential

When to use: Complex analysis requiring both data processing and judgement; creative work that benefits from AI-generated options with human curation; strategic planning with data-intensive inputs.

Governance consideration: Decision authority must be clear even in peer collaboration. When human and agent disagree, the resolution protocol must be defined in advance.

2.4 Escalation¶

Pattern: An agent encounters a situation outside its authority or capability and transfers control to a human (or to a higher-authority agent).

Characteristics: - Triggered by defined conditions (confidence threshold, authority boundary, novel situation) - Agent provides full context to the escalation target - Control transfers cleanly — no ambiguity about who owns the decision - Post-escalation, the agent may resume execution based on the human's decision - Escalation events are logged and analysed for pattern improvement

When to use: Built into every autonomous and orchestrator agent as a fundamental capability. This is not a failure mode — it is a design feature.

Governance consideration: Escalation triggers must be defined at design time. Escalation response SLAs must be set. Chronic escalation patterns should trigger process or authority redesign.

2.5 Agent-to-Agent Coordination¶

Pattern: Agents coordinate with each other to complete work that spans multiple agent capabilities, with or without human involvement.

Characteristics: - Agents invoke, inform, and hand off to other agents - Shared protocols for state transfer and context passing - May form agent chains or agent networks for complex tasks - Human visibility into agent-to-agent coordination is maintained - An orchestrator agent or human serves as the coordination authority

When to use: Complex automated workflows; multi-domain tasks; high-volume processing pipelines; situations where human coordination would be a bottleneck.

Governance consideration: Agent-to-agent coordination must be architecturally visible — not hidden. The full chain of agent interactions must be auditable. Human intervention points must be defined even in fully automated chains.

2.6 Human Override¶

Pattern: A human intervenes to stop, redirect, or take over agent activity regardless of the agent's current state or autonomy level.

Characteristics: - Universal capability — applies to all agent types and all situations - Cannot be refused, delayed, or circumvented by the agent - Override is logged with full context (who, when, why, what state) - Agent gracefully transfers state to the overriding human - Post-override recovery procedures are defined

When to use: Emergency situations; detected agent errors; regulatory requirements; trust recalibration events; any situation where a human determines that agent activity should cease.

Governance consideration: This is a fundamental principle, not a pattern. Override capability must be designed into every agent from inception. Override authority must be defined (who can override which agents).

Collaboration Pattern Summary¶

Pattern	Direction	Autonomy Level	Primary Use
Delegation	Human → Agent	High	Task assignment
Supervision	Agent → Human (review)	Medium	Quality assurance
Peer Collaboration	Bidirectional	Shared	Complex problem-solving
Escalation	Agent → Human	Triggered	Exception handling
Agent-to-Agent	Agent ↔ Agent	Variable	Automated workflows
Human Override	Human → Agent (interrupt)	Revoked	Emergency/control

Part 3: Trust and Autonomy Levels¶

Trust between humans and agents is not binary. ZORBA defines a five-level Trust and Autonomy Scale that governs how much independence an agent has in any given context.

The ZORBA Autonomy Scale¶

Level	Name	Description	Human Involvement
A0	Inert	Agent has no authority to act. Provides information only when queried.	Human performs all actions
A1	Supervised	Agent performs actions, but every output requires human approval before taking effect.	Human reviews 100% of outputs
A2	Guided	Agent performs actions autonomously for routine cases; flags exceptions and edge cases for human review.	Human reviews exceptions + periodic sampling
A3	Trusted	Agent operates autonomously within defined boundaries. Human reviews by exception only. Agent self-monitors and self-escalates.	Human reviews escalations and periodic audits
A4	Autonomous	Agent operates with full authority within its domain. Self-governing within defined constraints. Human oversight is structural (audit, policy) rather than operational.	Human sets policy and reviews systemic performance

Trust Calibration¶

Autonomy levels are not static. They are calibrated based on:

Performance history — Agents that demonstrate consistent quality earn higher trust
Domain risk — Higher-risk domains warrant lower autonomy levels
Regulatory requirements — Some domains have mandated human oversight levels
Organisational maturity — Organisations new to agentic operations start lower
Incident history — Trust can be recalibrated downward after failures

Trust Calibration Protocol¶

1. Initial deployment: Agent starts at A1 (Supervised)
2. Performance review period: Defined duration with 100% human review
3. Trust assessment: Performance metrics, error rates, escalation quality
4. Promotion decision: Human authority required to increase autonomy level
5. Continuous monitoring: Ongoing performance tracking with automated alerts
6. Recalibration trigger: Defined conditions that cause trust level review
   (error spike, novel situation class, regulatory change, incident)
7. Demotion protocol: Process for reducing autonomy with defined recovery path

Autonomy Level by ZORBA Layer¶

Different layers have different maximum appropriate autonomy levels:

Layer	Typical Maximum Autonomy	Rationale
L1: Strategy	A0 (Inert)	Strategic decisions require human authority
L2: Objectives	A1 (Supervised)	Objective-setting requires human judgement
L3: Capabilities	A2 (Guided)	Capability design requires human oversight with agent input
L4: Processes	A3 (Trusted)	Process orchestration can be highly automated
L5: Activities	A4 (Autonomous)	Routine activities can be fully autonomous
L6: Work	A4 (Autonomous)	Work execution at scale requires agent autonomy

These are defaults, not mandates. A highly regulated industry may require A2 maximum at L5. A mature, data-rich operation may justify A3 at L3. The point is that the decision is made explicitly.

Part 4: Workforce Composition Design¶

Workforce composition — the blend of human and agent capabilities across the enterprise — is an architectural decision. ZORBA provides a structured approach.

Composition Assessment¶

For each capability, process, or activity, assess:

Factor	Question
Repeatability	Is this work consistent and pattern-based, or novel each time?
Data intensity	Does this work primarily involve processing structured data?
Judgement complexity	Does this require nuanced human judgement?
Speed requirement	Does the required speed exceed human capacity?
Scale requirement	Does the required volume exceed human capacity?
Relationship dependency	Does this require human relationships or empathy?
Regulatory mandate	Do regulations require human involvement?
Error consequence	What is the impact of errors — reversible or catastrophic?
Creative requirement	Does this require genuine creative originality?

Composition Profiles¶

Based on assessment, each function receives a composition profile:

Human-Essential (H): Work that must be performed by humans due to judgement, creativity, regulatory, or relationship requirements. Agents support but do not perform.
Human-Led, Agent-Supported (H+a): Human performs the primary work with agent assistance — preparation, analysis, follow-up.
Blended (H=A): Genuinely shared between humans and agents, with clear division of responsibilities within the function.
Agent-Led, Human-Supervised (h+A): Agent performs primary work with human oversight — review, exception handling, quality assurance.
Agent-Essential (A): Work that is optimally or necessarily performed by agents due to speed, scale, data intensity, or consistency requirements. Humans govern but do not perform.

Composition Evolution¶

Workforce composition is not static. ZORBA expects and enables composition evolution over time:

Typical evolution path:
H → H+a → H=A → h+A → A

As agent capabilities mature and trust is calibrated,
functions naturally shift toward greater agent involvement.
This shift must be deliberate, governed, and reversible.

Critical principle: The path can go in either direction. If an agent-led function experiences quality degradation, trust recalibration may shift it back toward human-led. Composition is a continuous design decision, not a one-way migration.

Putting It Together¶

The Blended Workforce Model gives organisations a practical vocabulary and structure for answering the question ZORBA poses: where does the human end and the agent begin?

The answer is: it depends — on the agent type, the collaboration pattern, the trust level, the composition profile, the domain, the risk, and the regulatory context. But with ZORBA, it depends on deliberate architectural decisions, not on whatever happens to emerge when someone connects an AI to a business process.

The organisations that thrive in the agentic era will be those that design their blended workforce intentionally. ZORBA provides the blueprint.

Previous: ← Framework Architecture | Next: Domain Reference Model →