Ivory & Co.
AEA by Ivory & Co.StandardsKnowledge & ResearchImplementationCertificationIvory & Co.

The hybrid enterprise in AEA by Ivory & Co.When the workforce is no longer entirely human.

A hybrid enterprise fulfils its work through people, AI agents, technology and outside organisations in combination. Designing one means keeping capability, authority and responsibility as separate questions.

An AI agent receives a customer enquiry. The customer has been waiting three weeks for a refund.

The agent reviews the order, checks the transaction history, identifies the delay and determines that the customer is entitled to their money. So far everything is working beautifully.

The agent issues the refund, adds a goodwill credit and sends an apology. The customer is delighted.

There is one problem. The agent was never authorised to issue goodwill credits. It had the technical access. It understood the situation. Its choice may even have been commercially sensible. Somewhere between the ability to perform an action and the authority to take it, the enterprise lost control of a decision.

This is one of the defining design challenges of the hybrid enterprise. As people, AI agents, technology and outside organisations all take part in fulfilling what a business requires, the business has to become more precise about the work being done, who or what is doing it, and the conditions under which their actions are permitted and relied upon.

AI agents and the changing nature of participation

For generations, organisations have been designed around familiar assumptions. People perform work. Managers direct them. Technology supports them. External providers deliver services under contract. Those assumptions shaped organisational charts, job descriptions, performance management, procedures and governance.

They also shaped how businesses think about capacity. When demand rises, we consider hiring, improving productivity, outsourcing, or investing in systems that make existing employees more efficient.

AI agents add another possibility: technology that takes part in the work directly. An agent can interpret a request, retrieve information, determine the next action, coordinate with other systems and complete an activity with little human intervention. Several agents may work together. People may supervise them, collaborate with them or step in only under particular conditions. A supplier may provide human and agentic services inside the same commercial arrangement.

Participation can now take many forms, and those forms can change while the purpose of the business stays the same. An enterprise therefore has to be understood independently of assumptions about who, or what, performs its work.

Start with the work

Consider a business that processes insurance claims. Employees receive claims, review supporting information, assess eligibility, determine outcomes and communicate decisions. Some claims are simple. Others involve complicated circumstances, incomplete evidence or large sums.

Now the business introduces AI agents. One receives and classifies claims. Another gathers information. A third evaluates whether the evidence satisfies established criteria. A human specialist reviews cases that need professional judgement or additional authority.

The distribution of work has changed substantially. The requirements of the work have not gone anywhere. What information is needed? What must be established before an action can occur? Which activities need specialist competence? What conditions must be met before an outcome can be accepted?

Those questions exist whoever performs the activity, and they show where each form of participation fits. An agent may be excellent at gathering information and unsuitable for certain consequential determinations. A person may have the expertise and lack the authority to approve an outcome. An outside specialist may fulfil a requirement without belonging to the organisation at all.

This asks for far more precision than sorting work into tasks people do and tasks machines can automate.

Capability does not establish authority

Return to the agent that issued the goodwill credit. Its action raises six separate questions.

Was the agent capable of identifying an appropriate resolution? Did it have access to the information required? Was it permitted to issue the refund? Was it authorised to offer additional compensation? Could the business rely on its assessment? Who was responsible for setting and maintaining the conditions under which it operated?

The questions are related, and their answers can all differ.

The same is true of people. An experienced employee may know exactly how to resolve a commercial dispute and still need approval before changing contractual terms. A junior employee may be able to see sensitive information without being authorised to disclose it. A manager may hold formal authority and need specialist advice before using it responsibly.

AI makes these distinctions more visible, because technical capability now reaches well past the boundaries businesses have traditionally associated with software. An agent may be able to do something long before the enterprise has decided whether it should. Technical access makes an action possible. Permission is a separate matter.

Who answers for the outcome?

An AI agent prepares supplier payments. It gathers invoices, checks purchase orders, identifies discrepancies and assembles a payment batch. The business has designed the arrangement so that an authorised person approves the final release.

One morning the agent misclassifies an invoice. The error passes through and the supplier is overpaid. Who is responsible?

Several responsibilities are in play. The agent’s assigned activity was to prepare the payment information correctly. The person approving the batch had responsibilities attached to that approval. The business was responsible for the design, operation, supervision and assurance of the arrangement. A technology provider may have obligations under its contract.

Working this out takes more than identifying the last participant to touch the transaction. Replacing an employee with an agent does not transfer every responsibility that employee carried. Placing a person somewhere in the workflow does not settle accountability either. The arrangement has to be designed deliberately, and much of the commercial value of automation depends on how well the enterprise does that.

The organisational chart tells part of the story

Picture a business with twenty employees. A managing director oversees sales, operations and finance, each with a small team.

Now consider how it actually operates. A marketing agency manages demand generation. A logistics provider handles distribution. An accounting firm provides financial services. AI agents process customer enquiries, support purchasing decisions and prepare operational reporting. Several technology platforms coordinate all of it.

The business employs twenty people. The enterprise runs on far more than twenty participants.

This matters when leadership considers change. Expanding into a new market may call for new capability without a matching rise in permanent staff. Bringing an outsourced activity in-house changes commercial relationships, participation, technology and dependencies together. Automating a function may reduce some human work and increase the need for different skills, supervision and assurance. An enterprise that can only see its own employees is managing part of itself.

How a hybrid workforce of people and AI agents changes the economics of scale

Businesses have historically grown capacity by adding resources. More customers needed more service staff. More transactions needed more administration. More complexity needed more management.

Technology has been loosening those relationships for decades, and AI agents may loosen them a good deal further. A small business may be able to reach operating capacity that once required a much larger workforce. An enterprise may distribute work across people, agents and providers according to what each activity requires.

These are possibilities, and the economics are more involved than swapping labour costs for software costs. Agents must be enabled and maintained. Information has to be available and fit for use. Actions may need monitoring, review or escalation. Exceptions need handling. Dependence on outside platforms, models and providers has commercial consequences of its own.

How much of a workforce should be human has no universal answer. A business with complex physical operations has different requirements from a digital service provider. A regulated activity carries conditions a less constrained business does not face. A proposition built on personal relationships may deliberately keep people at its centre.

The future workforce is a design decision.

What happens when the participant changes?

A retailer employs a purchasing officer who reviews inventory, assesses supplier offers and prepares replenishment orders. The business decides to introduce an AI agent to do much of that work.

Consider what the purchasing officer actually contributes. They understand seasonal demand, supplier reliability, product substitutions, commercial terms and exceptions that existing systems represent poorly. They exercise judgement within purchasing limits. They maintain relationships that shape how suppliers respond when something goes wrong.

Some of that can be reproduced through technology. Some needs additional information, different arrangements or continued human involvement. The business has to determine which requirements remain, which have changed and how the new participant can satisfy them.

It also has to understand the dependencies it has created. An agent relying on inaccurate inventory information will make poor recommendations with remarkable consistency. If supplier terms change, its instructions need revisiting. If it becomes unavailable, the business needs another way to keep operating. And if it performs well, the business may find it can redesign replenishment altogether.

The participant has changed, and the consequences reach well beyond the participant. Intelligent systems can only navigate what has been described. AI does not replace the need for architecture. It exposes the absence of it.

People and agents will not always do the same kind of work

It is tempting to picture a simple division of labour, with agents on routine activity and people on complex activity. That describes some arrangements today and is unlikely to hold.

Agents may take on increasingly complex analytical, administrative and operational work. People may continue with routine work wherever physical presence, customer preference, professional requirements or commercial sense make that appropriate. Outside organisations may supply specialist capability through combinations of both. The distribution will vary and will keep moving.

That is a challenge for any business that defines work mainly through job titles. A job description bundles several requirements into one position: specialist knowledge, coordination, judgement, communication, administration. Once different forms of participation are available, those bundles may no longer be the best way to meet the requirements inside them. Examining them may produce new roles and new arrangements. It may equally confirm that an existing arrangement is still the best available.

Designing for change without losing control

A hybrid enterprise cannot be designed once and left. Models are updated. Suppliers alter their services. Regulations and customer expectations shift.

Imagine an agent originally authorised to answer product questions. Over time its abilities improve. It becomes able to recommend alternatives, negotiate certain conditions and initiate transactions. The enterprise may want to use those abilities, and greater technical ability does not by itself justify a wider remit.

What new work could the agent perform? What information would it need? What authority would be appropriate? What conditions must exist before its actions can be relied upon?

An adaptive enterprise can examine changes like these while keeping a coherent account of its current arrangements. It knows what has been designed, what has been authorised, what is operating and what remains a possibility.

The enterprise beyond its employees

For a long time, the size of an organisation has served as a convenient indicator of its capacity. Headcount, departments, management layers and premises are visible evidence of scale. They will keep mattering, and they may become less reliable guides to what an enterprise can accomplish.

A small organisation may coordinate substantial capability through agents, technology and outside relationships. A large one may discover that much of its complexity comes from arrangements it no longer needs. The relationship between organisational size and enterprise capability is becoming more open to design.

The businesses best placed to explore that will understand their requirements, the work needed to meet them and the conditions under which different participants can operate.

If people are no longer the only participants capable of performing enterprise work, how should we design the business?

The formal treatment of roles, participants and authority is in the Workforce Architecture Standard.

Keep readingEvery decision creates a different enterprise.Knowledge and research · Consequences →
→Continue

Where to go next.