AI is beginning to change more than productivity. It is changing how work itself gets performed. That creates a leadership challenge across software, medicine, finance, law, consulting and other professions: people responsible for redesigning work may need firsthand experience with the new AI-driven methods before they can effectively decide how everyone else should use them.
THE SIGNAL
Leadership traditionally creates leverage by moving away from individual production.
The junior engineer writes code.
The staff engineer designs systems and guides teams.
The physician becomes a department chair.
The lawyer becomes a partner.
The analyst becomes a managing director.
The teacher becomes an administrator.
As professionals become more senior, they often spend less time performing the underlying work and more time managing people, allocating resources, setting strategy and making higher-level decisions.
That progression makes sense when the underlying method of doing the work is relatively stable.
Artificial intelligence may be creating an exception.
AI is not merely making existing tasks faster.
Agents are beginning to change how tasks are assigned, researched, completed, reviewed, coordinated and even conceived.
That means the production process itself is being rewritten.
And during a period when nobody completely understands what the new production process should look like, leaders may not be able to rely entirely on what they learned earlier in their careers.
They may have to relearn the work.
WHAT THE MARKET IS MISSING
Much of the conversation around AI leadership focuses on strategy.
Executives are told they need an AI strategy.
Companies need AI governance.
Employees need AI training.
Organizations need responsible-use policies.
All of that matters.
But there is a more fundamental question:
How do leaders design an AI-native workflow if they have never personally experienced one?
Consider software engineering.
A senior technical leader who built software for 20 years may understand architecture, databases, testing, security and engineering organizations extraordinarily well.
But agentic coding introduces new questions.
How much context should an agent receive?
When should one agent become five?
When does autonomy improve productivity?
When does it create technical debt?
How should humans review code that humans didn’t write?
How should a repository be structured so agents can understand it?
How much work can an agent perform before human intervention becomes necessary?
These questions are difficult to answer entirely from dashboards and presentations.
They require experimentation.
That same issue is beginning to appear outside engineering.
OpenAI reported in June that agentic tools were spreading rapidly beyond developers into legal, recruiting and other nontechnical departments, with workers increasingly delegating substantive, multi-step tasks rather than simply asking chatbots questions. More recent enterprise data shows Codex adoption growing far faster in areas including legal, sales, recruiting and marketing than in engineering itself. (OpenAI)
This means software engineering may simply be where the transition became visible first.
The underlying leadership problem could spread almost everywhere.
FIRST-ORDER EFFECTS
Leaders may need to become practitioners again
Not permanently.
And not because leadership suddenly stops being about leverage.
But during the transition, senior professionals may need direct experience with AI tools in order to understand what those tools actually change.
A law-firm partner deciding how associates should use AI may need to personally examine how an AI system conducts research, drafts arguments, analyzes documents and makes mistakes.
A hospital leader evaluating clinical AI may need to understand how the system fits into an actual physician workflow rather than simply reviewing accuracy percentages.
A finance executive redesigning an analysis process may need to observe what happens when an AI agent researches, calculates, reconciles and escalates decisions.
The American Medical Association has explicitly emphasized involving physicians in the design, evaluation and implementation of AI-enabled clinical workflows, arguing that doctors should help shape how the technology fits into actual care delivery. (American Medical Association)
Harvard Law School is making a similar move. Its 2026 AI programs now include hands-on use of AI tools, not simply lectures about their strategic or legal implications. (Harvard Law School)
That is telling.
The people expected to govern these systems increasingly need experience inside the systems.
AI literacy moves upward
Corporate AI training has often focused on ordinary employees.
That may be backwards.
Senior leaders may need just as much education — possibly more — because they are deciding which workflows will change.
CIBC, for example, began its AI transformation with deep, hands-on AI immersions for its board and senior leaders, followed by full-day training for executives before expanding training across the wider organization. (Microsoft)
That suggests an emerging leadership principle:
AI fluency cannot stop at the management layer.
Experimentation becomes part of leadership
Traditionally, experimentation could often be delegated downward.
AI makes that more complicated.
If the purpose of experimentation is merely to determine whether a tool works, delegation is fine.
But if the purpose is to develop the judgment necessary to redesign an entire profession’s workflow, leaders may need firsthand exposure.
Someone else can test the tool.
They cannot completely outsource the judgment gained from using it.
SECOND-ORDER EFFECTS
This is where the signal becomes much larger.
The distance between leadership and work may shrink
Organizations traditionally develop layers.
Executives set direction.
Managers translate strategy.
Workers execute.
AI could compress parts of that structure.
A senior leader equipped with powerful agents may be able to interrogate data directly, prototype ideas, analyze documents, model decisions or build working demonstrations without waiting for several organizational layers to perform the work.
That doesn’t necessarily eliminate management.
But it could make senior professionals closer to production again.
The future executive may simultaneously be more strategic and more hands-on.
Not because they personally perform every task.
Because AI makes it possible for them to interact directly with the underlying work at much lower cost.
Experience could become less durable
Professional expertise traditionally compounds.
Twenty years of experience generally means twenty years of accumulated judgment.
AI introduces an uncomfortable possibility:
Some operating knowledge may depreciate faster.
A leader can remain extremely knowledgeable about medicine, law, finance or engineering while becoming increasingly unfamiliar with how younger professionals actually perform those jobs using AI.
That creates two different kinds of expertise:
Domain expertise — understanding the profession.
Production expertise — understanding how the profession is currently practiced.
Historically, those moved together.
AI could pull them apart.
A senior lawyer may understand law better than almost anyone in the firm while understanding an AI-native legal workflow less well than a third-year associate.
A hospital executive may understand medicine deeply while having little experience working beside an ambient clinical agent.
A senior software architect may understand computing systems extremely well while rarely using autonomous coding agents.
That does not make the senior professional less valuable.
But it could create a leadership blind spot.
Junior employees may temporarily know something senior employees don’t
This could create an unusual reversal.
Younger workers normally learn practices from senior professionals.
During a technological transition, knowledge can temporarily flow upward.
OpenAI’s recent enterprise data found AI usage highest among early-career employees and declining among more senior employees. (OpenAI)
That matters.
The junior employee still needs the senior professional’s judgment.
But the senior professional may increasingly need the junior employee’s familiarity with new production methods.
The strongest organizations may therefore become more bidirectional:
Senior people teach judgment.
Junior people expose senior people to emerging workflows.
AI sits between them.
Leadership credibility could change
Employees are more likely to trust guidance about a technology when the person giving that guidance understands how it actually behaves.
Imagine being told:
“You should use AI for this.”
by someone who has never attempted the task with AI.
Or:
“AI can handle that.”
by someone who has never experienced the system fail.
Or:
“Humans must review this.”
without understanding which portions actually require human judgment.
During stable periods, leaders can manage through abstraction.
During technological transitions, credibility may increasingly come from understanding the reality underneath the abstraction.
WINNERS
Hands-on leaders
The winners won’t necessarily be leaders who use AI the most.
They will be leaders who understand it well enough to distinguish capability from hype.
They will know what should be automated.
What shouldn’t.
Where agents require supervision.
Where employees are wasting time.
Where AI changes the entire workflow rather than merely accelerating one task.
Organizations that redesign work
Deloitte’s 2026 research found that 48% of organizations had introduced AI without redesigning the workflows or roles around it, while only 12% reported redesigning at scale.
That is an enormous gap. (Deloitte)
The likely advantage goes to organizations that don’t simply give workers AI.
They reconsider:
What should the job now be?
Which steps disappear?
Which decisions remain human?
Which responsibilities move to agents?
Where does human judgment become more valuable rather than less?
That requires leaders who understand both the existing work and the emerging alternative.
Professionals who combine experience with experimentation
The most valuable senior employees may not be the people who know the old system best.
Nor simply the people who know the newest AI tools.
They may be the people who can combine:
deep domain experience + AI experimentation + organizational judgment.
That combination could become exceptionally valuable.
Companies that allow safe experimentation
KPMG’s 2026 research into finance found that one of the largest barriers to developing AI capability was a lack of hands-on practice environments, cited by 61% of respondents.
At the same time, 93% of U.S. companies surveyed expected to be deploying or scaling AI across finance within the following 18 months. (KPMG)
That gap matters.
You cannot create experienced AI leaders without allowing people to actually use AI.
LOSERS
Leaders who delegate all AI understanding downward
Delegating implementation makes sense.
Delegating experimentation can make sense.
Delegating all understanding of how the work is changing is more dangerous.
Executives could find themselves approving strategies built on assumptions that no longer match how employees actually operate.
Organizations that mistake tool deployment for transformation
Buying thousands of AI licenses is easy.
Changing the work is hard.
Microsoft, Deloitte, McKinsey and others are increasingly converging around the same point: meaningful AI value comes from redesigning workflows and operating models rather than simply placing AI on top of existing processes. (Microsoft Blogs)
An organization can therefore look highly advanced by traditional metrics —
thousands of users,
millions of prompts,
hundreds of agents —
while leaving the underlying work almost unchanged.
Leaders who confuse AI activity with competence
There is an equally dangerous mistake in the opposite direction.
A leader using AI constantly is not automatically a good leader.
Token consumption isn’t judgment.
Prompts aren’t strategy.
Lines of AI-generated code aren’t business impact.
The objective is not to turn executives into the heaviest AI users in the company.
The objective is to ensure they have enough firsthand experience to make informed decisions about a changing production system.
Companies that freeze today’s AI workflow too early
No one yet knows the final form of AI-powered work.
Organizations that establish rigid procedures too quickly could lock themselves into methods designed around today’s models.
The technology is still moving too rapidly.
Leadership therefore needs not only an AI operating model.
It needs the ability to continuously relearn the operating model.
WHAT HAPPENS NEXT
The first phase of enterprise AI was about access.
Give employees AI.
The second phase was about adoption.
Get employees to use AI.
The next phase appears to be about redesign.
Change how the work gets done.
That transition will put increasing pressure on leadership.
Expect more companies to create executive AI immersions, internal laboratories, hands-on training, cross-functional experimentation groups and safe environments where leaders can personally test emerging workflows.
Professional schools may change too.
Harvard Law’s move toward hands-on AI learning is probably an early indicator.
Medical education will increasingly need to address AI-supported clinical practice.
Business schools will need to teach management of human-agent organizations.
Engineering leadership programs may need to teach orchestration rather than simply architecture.
The line between learning the technology and leading the organization will become increasingly difficult to separate.
And this will spread far beyond technology companies.
In healthcare, McKinsey argues that AI’s real value requires redesigning care and operational workflows rather than adding isolated tools. Its nursing research similarly found that transformation requires rethinking how work is distributed among nurses, care teams and digital systems. (McKinsey & Company)
In finance, AI agents are moving into research, treasury, wealth management, onboarding and other operational workflows while institutions maintain human oversight around critical decisions. (Reuters)
In consulting, BCG is training employees across backgrounds not only in general AI use but increasingly in technical capabilities including building AI agents. (Business Insider)
The pattern is becoming visible.
AI expertise is moving into the professions themselves.
Eventually standards will stabilize.
Best practices will emerge.
AI interfaces will mature.
At that point, leaders may once again be able to move farther away from direct experimentation.
But we may not be at that point yet.
We may be living through the period before the new rules have been written.
And the people responsible for writing those rules may need to experience the new work firsthand.
BOTTOM LINE
Leadership has always been about leverage.
The best leaders do not produce everything themselves.
They create systems that enable many other people to produce more.
AI does not change that principle.
It may change what leaders need to understand before they can create that leverage.
When technology simply improves an existing tool, leadership can often remain above the workflow.
When technology changes the workflow itself, that distance becomes dangerous.
Software engineering is providing an early example.
Senior technical leaders who once gained leverage by stepping away from day-to-day coding may temporarily need to step closer again because coding agents are changing the production process they are responsible for designing.
But engineering is probably only the beginning.
Doctors will help determine how AI participates in care.
Lawyers will determine how agents participate in legal work.
Financial leaders will determine where AI can analyze, recommend and act.
Educators will determine what learning looks like when every student has access to machine intelligence.
Executives will determine how humans and agents divide responsibilities across entire organizations.
Those decisions cannot be based solely on what AI is supposed to do.
Leaders need to understand what happens when people actually work with it.
The leadership principle of the AI transition may therefore become surprisingly simple:
You don’t have to do everyone’s job.
But when technology changes how the job is done, you may have to learn the job again.
