The Next AI Advantage Will Be Management, Not More Agents

For the past few years, the enterprise AI conversation has focused on adoption.

How many employees are using AI? How many copilots have been deployed? How many pilots are running? How many agents have moved into production?

Those questions made sense when companies were still experimenting.

They are becoming less useful now.

As AI agents begin to take on meaningful work across finance, HR, customer service, marketing, engineering, and operations, the management challenge changes. Organizations are no longer simply introducing new tools. They are beginning to introduce a new category of worker into the operating model.

That raises a much more important question:

Who is managing the digital workforce?

McKinsey’s recent discussion, Your AI agents need performance management, too, gets to the heart of this issue. The central message is not really about agents. It is about the way organizations need to rethink performance, accountability, talent, and work itself as humans and AI increasingly operate side by side.

For leaders, this is where the AI conversation needs to mature.

Technology is not the transformation

One of the strongest points in the McKinsey discussion is that the technology is often the easier part.

Organizations can buy software, connect models, build agents, and automate workflows. The harder question is whether the business itself has the capacity to absorb the change.

That distinction matters.

Too many companies still approach AI by adding a new technology layer on top of an old process. The workflow remains the same, the decision rights remain the same, and the organization remains the same. AI simply makes parts of it faster.

That is not transformation.

If the process is badly designed, AI may accelerate the bad process. If accountability is unclear, automation can make it even harder to understand who owns the outcome. If the data is fragmented, AI can simply consume fragmented information faster.

This is why I keep coming back to one principle:

You do not transform a business by adding AI to yesterday’s processes.

The real opportunity is to reconsider how the work should be done when technology can take on more of the execution.

That requires business leadership, not just technology leadership.

Every agent needs an owner

As agents proliferate, one of the first disciplines companies will need is clear ownership.

It would be easy to assume that agents belong to the CIO, CTO, or AI team because they are technical systems.

I think that would be a mistake.

If an AI agent is performing work in finance, finance should remain accountable for the outcome. If it supports HR, HR should own the business performance. If it is working inside customer service, the customer service leadership should remain responsible.

Technology should provide the architecture, platforms, security, engineering standards, and guardrails.

But business accountability should not move simply because part of the work is now being performed by a machine.

A useful principle for the agentic enterprise is therefore:

No production agent without a clear business owner.

That owner should know what the agent is supposed to do, what systems and data it can access, how its performance is measured, and what happens when it gets something wrong.

Without that discipline, companies risk creating a digital workforce faster than they create the ability to govern it.

Deployment is not performance

This is where many AI programs still get distracted by the wrong metrics.

Adoption matters, but adoption is not value.

An organization can have thousands of active users and dozens of AI agents while producing very little measurable business impact.

The more important questions are operational.

Has the finance team reduced closing time?

Has customer service improved resolution quality?

Has engineering reduced cycle time?

Has marketing improved conversion or lowered production cost?

Has decision quality improved?

Has risk declined?

Has revenue increased?

Those are the metrics that matter.

An agent should not be considered successful because it exists, or even because people use it.

It should be considered successful because it improves the business outcome it was designed to support.

That sounds obvious, but enterprise technology has a long history of confusing implementation with success.

AI cannot afford to repeat that mistake.

Agents need a lifecycle

McKinsey also raises an important point around lifecycle management.

We understand how to manage people over time. Employees are hired for a role, onboarded, evaluated, developed, reassigned, and eventually leave or retire.

Agents need something similar.

An agent that works well today may not remain appropriate forever. The model may change. The business process may evolve. Costs may increase. Another system may perform the same task better. Its instructions may become outdated. Its permissions may no longer be appropriate.

McKinsey even uses the concept of “abandonware” to describe technology that remains inside the organization after people have effectively forgotten about it.

That risk becomes more serious as agent numbers increase.

A company with ten agents can probably manage them informally.

A company with hundreds or thousands cannot.

At some point, organizations will need a proper inventory of their digital workforce. They will need to know which agents exist, who owns them, what they cost, what they access, what value they produce, and when they were last reviewed.

This is not about adding bureaucracy.

It is about making autonomy manageable.

AI economics will become a CFO issue

There is also an economic dimension that leadership teams should not underestimate.

Pilots often look cheap.

Then they succeed.

Usage grows, more employees use the system, agents make more model calls, consume more compute, retrieve more data, and interact with more applications.

Suddenly the economics look different.

McKinsey makes the point that companies need to think earlier about architecture, token consumption, forecasting, and vendor relationships if they expect AI solutions to scale.

This is where the CFO needs to become much more involved.

The question should not simply be, “How much does this agent cost?”

It should be:

What value does this agent create relative to what it costs?

A more expensive AI interaction can make excellent economic sense if it creates a much larger business outcome. A cheap interaction repeated millions of times without meaningful value can still be wasteful.

The objective is not to minimize AI spend.

It is to maximize the return on AI-enabled work.

Your best AI talent may already work for you

One of the most useful observations in the report concerns domain expertise.

Companies often assume AI transformation means hiring more AI specialists.

Some of that talent is necessary.

But organizations should not underestimate the knowledge already inside the business.

McKinsey cites a memorable example: it can be easier to teach a metallurgist AI than to teach an AI specialist metallurgy.

Every company has its own equivalent.

The underwriter who understands risk. The engineer who understands the plant. The salesperson who understands the customer. The operations leader who knows why the process works differently from what the documentation says.

AI makes that knowledge more valuable, not less.

The strongest companies may not be those that hire the largest number of AI specialists.

They may be the companies that become best at combining technical capability with deep institutional knowledge.

That is why AI literacy across the business matters so much.

The human side of AI may become harder, not easier

One of the more counterintuitive points in the McKinsey discussion is the effect AI can have on cognitive load.

We often assume that AI will make work easier because it removes repetitive tasks.

And it can.

But when routine work disappears, what remains is often the difficult work.

Judgment. Decision making. Creativity. Ambiguity. Conflict. Problem solving. Human interaction.

McKinsey notes that intensive AI users can sometimes experience greater cognitive load precisely because AI is working.

That is an important warning for leaders.

Productivity should not mean removing every simple task and filling the entire working day with high-intensity judgment.

AI may reduce the amount of work while increasing the intensity of the work that remains.

That changes how organizations should think about workforce design, workload, decision fatigue, management, and performance.

The objective should be a better division of labor between humans and machines, not simply maximum utilization of both.

Management becomes the advantage

For the last several years, some of the competitive advantage in AI came from access to models, data, engineering talent, and computing capacity.

Those advantages still matter, but they are becoming less exclusive.

AI tools are becoming easier to use. Development barriers are falling. Capabilities that were once available only to sophisticated technology teams are moving into the hands of ordinary employees.

The source of advantage therefore begins to shift.

It moves from access to execution.

From experimentation to operating model.

From deployment to management.

The next chapter of enterprise AI will not be defined by which company has the largest number of agents.

It will be defined by which company knows how to manage them.

The strongest organizations will understand which agents should exist, what outcome each one supports, who owns them, what they cost, what risks they create, how their performance is measured, and when they should be improved or retired.

Just as importantly, they will understand what those agents change for the humans working alongside them.

That is the real transformation.

AI agents are entering the workforce. The leadership challenge is no longer simply to deploy them faster.

It is to build an organization in which human and machine capabilities are combined deliberately, governed clearly, and measured by the value they create.

AI is becoming a management capability.

And the companies that understand that early will have a significant advantage.

---

Source: McKinsey & Company, People & Organizational Performance Practice, “Your AI agents need performance management, too,” August 2026.