Let me start with a statement that may sound deliberately provocative.

If you are not already exploring how AI agents can work inside your business, you are falling behind.

I am not suggesting that every organisation that has not deployed agents is already obsolete. That would be an exaggeration. What is becoming obsolete, however, is the idea that every meaningful task must be completed manually, one step at a time, by a human employee.

That operating model belongs to a world that is disappearing faster than many leaders realise.

For the past few years, most conversations about artificial intelligence have focused on assistants. We ask AI to write an email, summarise a report, review a document, prepare a presentation or help us research a subject.

AI agents represent something different.

Instead of asking the system to complete a single task, we can give it an objective. The agent can then plan the work, divide it into steps, use different tools, retrieve information, interact with systems and return with a completed result, or ask for human intervention when necessary. The difference may appear subtle, but it is fundamental, as far as I can see.

Using an AI assistant is similar to using a sophisticated tool. Managing an AI agent is closer to delegating work to a junior employee.

Of course, the agent does not possess human judgment, genuine understanding or personal accountability (now). But it can increasingly execute work, rather than provide information.

This is why agents have become the latest focus of the technology industry. Almost every major software company is investing in them. Consultants are publishing increasingly ambitious forecasts. Investors are funding agentic platforms, and vendors are rapidly adding the word “agent” to their products.

Some of this is undoubtedly hype.

But hype does not necessarily mean that the underlying change is imaginary.

According to McKinsey’s 2025 global AI survey, 88% of respondents said their organisations were using AI regularly in at least one business function.

More importantly, 62% said their organisations were already experimenting with AI agents. Twenty-three percent reported that they were scaling at least one agentic system somewhere within the business.

I think those numbers need to be interpreted carefully. Experimenting with an agent is not the same as transforming a company. McKinsey also found that, within individual business functions, relatively few organisations had managed to scale agents extensively. That is exactly why I believe the present moment matters.

The technology is advancing rapidly, but most organisations have not yet adapted their structures, roles or management systems around it.

This creates an opportunity for those willing to move early, but also a significant risk for those who continue waiting for the picture to become completely clear. Let me tell you, it will never become completely clear.

Technology does not normally arrive with a perfectly developed instruction manual, a proven organisational structure and a workforce already trained to use it. Companies learn by experimenting, governing, correcting and improving.

Those that begin today will accumulate experience.

Those that delay will eventually have to acquire the technology and the organisational knowledge at the same time, and here is where I believe the real disadvantage begins.

Productivity will no longer depend only on headcount

Imagine two competing companies: both employ approximately the same number of people. Both operate in the same market and have access to similar information.

In the first company, employees complete research, analysis, reporting, administration and follow-up activities manually.

In the second, employees use agents to monitor information, prepare preliminary analysis, update systems, draft communications and manage routine workflows. Humans remain responsible for the objectives, the critical decisions and the outcome, but agents provide additional execution capacity.

The second company may be able to review more opportunities, respond to more customers, analyse more data and test more ideas without increasing its headcount at the same rate.

Obviously, the difference may not be dramatic on the first day, but believe me, it may begin with an hour saved here and a report completed more quickly there; but multiplied across hundreds of employees and thousands of activities, those small improvements become an operating advantage.

Over time, one organisation learns how to orchestrate human and digital work together. The other continues to treat AI as an optional productivity tool. This is why I say that companies risk missing the bus.

The issue is not simply whether they own the latest software. It is whether they are learning how to operate in a world where execution is no longer limited exclusively by the availability of human labour.

Every employee becomes a manager

The most important consequence of this transformation is not technological. It is organisational.

In my view, every employee who uses AI agents will become a form of manager.

Not a manager in the traditional sense. They may have no employees reporting to them. They may not control a budget or hold a leadership title, but they will be managing work performed by non-human resources.

A financial analyst may manage an agent that monitors performance, investigates variances and prepares a first draft of the monthly commentary.

A salesperson may use agents to research potential clients, prepare meeting briefs, update the CRM system and draft follow-up communications.

An engineer may direct agents to review technical documentation, compare specifications, identify inconsistencies and prepare preliminary calculations for human verification.

A human-resources professional may supervise agents that support onboarding, prepare training material or answer routine employee questions.

These employees will no longer be responsible only for completing tasks themselves. They will have to decide which tasks to delegate, how to define the objective, what information the agent can access and how the result should be reviewed.

They will need to recognise when the agent has produced something useful, when it has misunderstood the assignment and when it has created a risk. That is management, or not?

Microsoft has used the expression “agent boss” to describe this emerging role. Its 2025 Work Trend Index found that many leaders expect employees to begin building, training and managing agents as part of their normal responsibilities.

Thirty-six per cent of leaders surveyed expected agent management to become part of their teams’ work, while 41% expected employees to be involved in training agents.

I find the expression interesting, although I am less interested in the title than in what it reveals.

Management is about to become distributed throughout the organisation. We are not prepared for this kind of manager.

For decades, companies have developed training programmes for people managers.

We teach delegation, communication, motivation, conflict resolution, performance management and leadership. These remain essential capabilities and will not disappear.

But managing an agent requires a different set of skills.

An employee must be able to describe the required outcome clearly. They need to provide enough context for the agent to work effectively without giving it unnecessary access to confidential information.

They must decide what the agent is allowed to do independently and where human approval is required.

They must evaluate the quality of the output, identify unsupported conclusions and question results that appear convincing but may be wrong.

They also need to understand when the workflow itself is poorly designed.

If an agent repeatedly produces incorrect or inconsistent results, the solution may not be to write a slightly better prompt. The problem may lie in the information provided, the systems it accesses, the permissions it has been given or the way the work has been divided between the human and the machine.

This is why I believe the current obsession with prompt engineering is too narrow.

Writing instructions is useful, but it is not the core capability companies need to develop.

The more important skill is orchestration: the ability to combine people, agents, information and technology into a controlled workflow that produces a reliable outcome.

That is a management capability, not merely a technical one.

Microsoft’s 2026 research found that AI users increasingly considered quality control and critical thinking to be essential human skills. Eighty-six percent of respondents said they treated AI output as a starting point rather than a final answer.

That is precisely the mindset organisations must encourage.

The best employee in the agentic workplace will not be the person who accepts everything produced by AI. It will be the person who knows how to extract value from AI while continuing to challenge, verify and improve its work.

Employees may be moving faster than their companies

One of my concerns is that many organisations are still treating AI as a technology deployment.

They purchase licences, nominate a few people to run pilot projects and perhaps issue a policy explaining what employees should or should not do. But the real transformation is far broader.

Microsoft’s 2026 Work Trend Index surveyed 20,000 AI users across ten countries. It found that only 19% were operating in what it described as the “Frontier” zone, where employee capability and organisational readiness supported each other.

Another 10% were already capable AI users but were being held back by organisations that had not created the necessary systems, incentives or support.

Only 26% said their leadership was clearly and consistently aligned on AI.

That finding does not surprise me; I see this every day talking with friends, managers, and board members.

Employees are already experimenting. They are finding ways to automate parts of their work, often without waiting for formal approval. Some are using publicly available tools. Others are quietly creating workflows that their managers may not even know exist.

The organisation frequently discovers the change only after it has happened, and this creates both an opportunity and a governance problem.

On one side, employees are demonstrating initiative and discovering useful applications. On the other, the company may have no visibility over what information is being shared, which tools are being used or how important decisions are being influenced.

The solution cannot simply be to prohibit the technology.

Employees will continue to use it because the productivity advantage is too obvious.

The organisation must instead create an environment in which experimentation can occur within defined boundaries, with appropriate support, training and accountability.

Job descriptions must change

Most job descriptions are still built around activities.

Prepare reports. Review documents. Contact clients. Update records. Coordinate meetings. Analyse performance.

But many of these activities can already be supported, and in some cases substantially executed, by AI agents.

Companies therefore need to reconsider how roles are defined.

Instead of asking, “Which tasks does this person perform?”, we should increasingly ask, “Which outcome is this person responsible for delivering?”

The distinction is important.

Tasks will change as technology changes. Accountability must remain clear.

A future employee may be responsible for an outcome that is produced through a combination of their own expertise, the support of several agents and the contribution of colleagues.

The individual’s value will not be measured by how many documents they personally typed or how many hours they spent collecting information. It will be measured by the quality of the outcome, the decisions they made and how effectively they managed the resources available to them.

This also challenges traditional methods of performance evaluation.

Activity becomes a weak measurement when an agent can create dozens of reports, emails or analyses in a short period of time.

A high volume of output does not automatically mean that useful work has been completed.

Performance management will need to focus more heavily on judgment, accuracy, outcomes, risk control, learning and the ability to improve workflows.

Traditional managers must become architects

The role of the formal manager will also change: a significant part of traditional management involves allocating work, coordinating resources and monitoring whether tasks have been completed.

When agents begin performing part of that execution, managers must move further towards the design of work.

They need to decide which processes should remain entirely human, which should be supported by agents and which may eventually operate with limited autonomy.

They must identify where human review is essential, what evidence should be retained and how exceptions should be escalated.

In other words, managers will increasingly become architects of operating systems.

This is one reason why buying an AI platform does not create an AI transformation.

McKinsey found that the organisations obtaining the greatest value from AI were much more likely to have fundamentally redesigned their workflows. They were also more likely to have strong senior-leadership ownership and clear processes defining when AI outputs required human validation. That is the part many organisations underestimate.

I keep screaming that technology can be purchased quickly. Organisational capability cannot.

It must be built through experimentation, leadership, governance and training.

Agents must be governed as digital workers

There is also another important difference between AI agents and the tools that preceded them.

A chatbot can provide a wrong answer.

An agent may be able to take the wrong action.

Depending on its permissions, an agent might access internal data, update a system, communicate with a customer, initiate a transaction or trigger another automated process.

The risk therefore extends beyond inaccurate content. It includes inappropriate execution.

Deloitte’s 2026 research surveyed more than 3,200 business and technology leaders involved in organisational AI programmes. Seventy-four percent expected their organisations to be using AI agents at least moderately by 2027.

Yet only 21% said they currently had a mature governance model for agentic AI. That is a serious gap!

Every agent should have a defined purpose, a responsible owner and controlled access to information and systems. Its actions should be monitored, and its performance should be reviewed.

There must be clear limits to what it can do without approval.

The organisation must also know when an agent should be modified, suspended or retired.

This may sound similar to the way we manage employees, suppliers or other business resources. In many respects, it is.

The important difference is that the human employee managing the agent must remain accountable for how it is used.

AI governance cannot remain a specialist activity performed only by IT, legal, compliance or risk teams. These functions must establish the framework, but governance must also operate through the daily behaviour of every employee using an agent.

That employee must understand not only what the technology is capable of doing, but what it is authorised to do.

Not every agent deserves to exist

I strongly believe that organisations must engage with agents now. But I do not believe that every process should be automated or that every agent project will create value.

Gartner has predicted that more than 40% of agentic AI projects may be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls.

It has also warned about “agent washing”, the practice of presenting ordinary chatbots or established automation tools as agents because the term currently attracts attention.

This is an important warning: urgency should not become recklessness.

The objective is not to deploy the largest possible number of agents. It is to identify where they can produce meaningful, measurable and controlled value.

Some organisations will fall behind because they refuse to engage with the technology.

Others may create different problems by deploying agents everywhere without understanding the cost, the risk or the operational purpose. Both approaches are failures of management.

A disciplined organisation should begin with clearly defined business problems. It should test agents in controlled environments, establish measurable outcomes and understand the potential consequences of failure. Only then should it scale.

The bus is still moving

I do not believe it is too late for companies that have not yet adopted agents.

But the window for treating AI as an interesting future possibility is closing.

Organisations should already be identifying where agents could create value, which jobs and workflows will be affected and what capabilities their employees will need.

They should be training people not simply to use AI, but to manage it.

They should also be establishing the governance structures required to ensure that delegation to an agent does not mean the disappearance of human accountability.

The employee of the future will not be valuable simply because they can complete a particular task.

They will be valuable because they can define an objective, deploy the right combination of human and digital resources, evaluate the result and remain responsible for the outcome. They may never receive a traditional management title.

Nobody may formally report to them. But they will still be managers.

The organisations that understand this will gradually develop management capability throughout their entire workforce.

Those that do not may continue to employ talented people and invest in sophisticated technology. They may even announce ambitious AI strategies.

But they will continue operating through structures designed for a world in which execution and intelligence were available only through human labour.

That world is changing.

The question to me is no longer whether AI agents will become part of the workforce.

The question is whether our organisations and our people will be ready to manage them.

References

McKinsey & Company — The State of AI in 2025
https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai

Microsoft — 2025 Work Trend Index: The Year the Frontier Firm Is Born
https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born

Microsoft — 2026 Work Trend Index: Agents, Human Agency, and the Opportunity for Every Organisation
https://www.microsoft.com/en-us/worklab/work-trend-index/agents-human-agency-and-the-opportunity-for-every-organization

Deloitte — Agentic AI Is Scaling Faster Than Guardrails
https://www.deloitte.com/us/en/insights/topics/emerging-technologies/ai-agents-scaling-faster.html

Gartner — Over 40% of Agentic AI Projects May Be Cancelled by 2027
https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027

World Economic Forum — The Future of Jobs Report 2025
https://www.weforum.org/publications/the-future-of-jobs-report-2025/digest/

This article is also published on Medium as part of my public research and writing.