Imagine two companies.

Both use exactly the same AI model.

Both have access to the same technology.

Both employ talented people.

Both have similar budgets.

Both are trying to automate the same process.

Yet one becomes dramatically more productive.

The other creates a mess.

Why?

It may have very little to do with the AI.

The difference may be one decision:

Who does what?

Which work belongs to the human?

Which belongs to AI?

Which can AI prepare but a human must approve?

Which can an agent execute independently?

And most importantly:

Who owns the outcome?

This may become one of the most important organizational questions of the AI era.

Because companies have spent the last few years asking:

“What can AI do?”

That question is becoming less useful every month.

AI can do more.

Agents can do more.

Models become more capable.

Automation becomes easier.

The important question is changing.

It is now:

“What should AI do?”

There is a big difference.

Capability Is Not Permission

An AI system might be capable of:

writing an employment offer,

evaluating a candidate,

approving a refund,

recommending a loan,

changing a price,

responding to a customer,

reviewing a contract,

diagnosing an anomaly,

selecting a supplier,

or recommending that an employee be promoted.

That does not mean it should independently make all of those decisions.

We make this distinction naturally with people.

A junior accountant may technically know how to transfer AED 2 million.

The company still doesn’t give them the authority to do it.

A salesperson may know the company’s pricing.

That doesn’t automatically give them permission to approve any discount.

A manager may understand an employee’s performance.

There may still be rules around dismissal.

Organizations separate:

Capability

from

Authority.

AI needs the same treatment.

Perhaps even more carefully.

Stop Asking “Human or AI?”

The debate is usually framed badly.

Will AI replace people?

Will humans still be necessary?

Should this job be automated?

These questions assume there are only two choices.

There aren’t.

The future of most knowledge work is likely to contain several operating modes.

A useful starting framework is:

Human Only

Human + AI

AI + Human Approval

AI Autonomous Within Limits

These are not stages every process must pass through.

They are different ways of allocating work.

And choosing between them requires more than asking whether AI is technically capable.

It requires thinking about:

judgment,

risk,

accountability,

trust,

reversibility,

economics,

and human value.

Mode 1 — Human Only

Let’s begin with the category some AI discussions try to ignore.

There will remain work where humans should own the decision.

Not necessarily because AI cannot help.

But because the human decision itself has value.

Consider a CEO deciding whether to close a business unit.

AI can analyze:

profitability,

market conditions,

forecasts,

employee costs,

strategic scenarios,

competitive dynamics.

Excellent.

But who should make the final decision?

The CEO.

Why?

Because the decision contains more than analysis.

It contains:

responsibility,

organizational consequences,

values,

trade-offs,

uncertainty,

and accountability.

Those are not simply computational issues.

Human Work Is Not Just “Whatever AI Can’t Do Yet”

This distinction matters enormously.

A terrible workforce strategy looks like this:

“Automate everything AI can do and leave whatever remains to people.”

That makes the human role accidental.

It guarantees that jobs become collections of:

exceptions,

problems,

approvals,

complaints,

and things the AI couldn’t figure out.

What a wonderful career.

A better organization intentionally asks:

Where does human contribution create disproportionate value?

That may include:

judgment,

trust,

relationships,

leadership,

negotiation,

taste,

ethical responsibility,

strategic intent,

creativity under ambiguity,

and accountability for consequential decisions.

Microsoft’s 2026 workplace research points in this direction: as AI and agents take on more execution, advanced users increasingly define human value around setting intent, applying judgment, maintaining standards and owning outcomes. [1]

That is very different from saying:

“Humans do whatever AI can’t.”

Consider a Difficult Customer

A customer has been with your company for 12 years.

Something went badly wrong.

They are angry.

There is significant money involved.

Technically, an AI agent may be able to:

read the entire customer history,

understand the complaint,

analyze company policy,

calculate compensation,

draft a response,

and recommend a resolution.

Useful.

Should the AI make the final call?

Maybe not.

Because the decision may require someone to recognize:

This customer matters strategically.

The policy doesn’t fit this particular situation.

The relationship is worth preserving.

The customer doesn’t need another perfect email.

They need a senior person to call.

That last judgment may be the most valuable action in the entire process.

AI should help the person arrive there faster.

It doesn’t necessarily need to replace them.

Mode 2 — Human + AI

This will probably become the default for a large share of professional work.

The person owns the outcome.

AI expands their capability.

An accountant investigates an unusual transaction with AI assistance.

A lawyer reviews arguments with AI.

A marketer develops positioning with AI.

A salesperson researches an account.

A doctor examines supporting information.

A manager prepares for a difficult conversation.

A consultant develops scenarios.

A designer explores alternatives.

The AI helps with:

search,

analysis,

synthesis,

comparison,

generation,

simulation,

and critique.

But the human remains deeply involved.

Microsoft’s 2026 Work Trend Index found that 49% of analyzed Microsoft 365 Copilot conversations supported cognitive work such as analysis, evaluation, problem-solving and creative thinking. It also found that 66% of surveyed AI users said AI allowed them to spend more time on higher-value work. [1]

This is an important reminder.

The future isn’t only automation.

AI can also increase the quality and reach of human judgment.

A Great Analyst With AI Is Not the Same as AI Without an Analyst

Consider financial analysis.

AI can quickly:

read financial statements,

calculate ratios,

identify trends,

compare periods,

summarize risks,

generate scenarios.

Does that mean the analyst disappears?

Sometimes routine analytical work may.

But expertise still matters.

The analyst knows:

which assumptions are strange,

which number deserves skepticism,

which trend is normal for this industry,

which explanation management always gives,

which risk isn’t visible in the numbers,

which question hasn’t been asked.

AI can dramatically expand the amount of information someone can process.

But access to more analysis does not automatically create better judgment.

This distinction becomes even more important as AI makes sophisticated-looking analysis available to almost everyone.

The output becomes cheap.

Knowing what deserves belief becomes expensive.

AI Should Often Be a Second Brain, Not the Final Signature

Human + AI is particularly powerful where:

the work is difficult,

AI can contribute meaningful analysis,

but context still matters.

Think about:

strategy,

leadership,

creative direction,

complex sales,

management,

advisory work,

medical decision support,

legal analysis,

investment analysis,

product design.

The AI can be:

researcher,

critic,

simulator,

analyst,

editor,

opponent,

assistant.

The person remains:

accountable.

And accountability is something companies should protect deliberately.

Mode 3 — AI Does the Work, Human Approves

Now we move closer to automation.

Imagine an insurance claim.

The AI:

reads the documents,

extracts information,

compares policy terms,

checks supporting evidence,

identifies inconsistencies,

calculates a recommended settlement,

and prepares the decision.

The human receives:

Recommended approval: AED 8,450.

Policy conditions satisfied.

Confidence: high.

No fraud indicators identified.

Evidence attached.

The human reviews and approves.

This can dramatically reduce the amount of work required without giving AI complete authority.

The same model could apply to:

purchase approvals,

refunds,

contracts,

pricing exceptions,

financial exceptions,

employee requests,

quality reviews,

customer escalations.

The important design question is:

What exactly is the human approving?

Human Approval Can Become Theatre

This is a dangerous problem.

Companies say:

“There is a human in the loop.”

That sounds reassuring.

But imagine an employee receives 400 AI decisions per day.

Each appears carefully analyzed.

The employee has 30 seconds to approve or reject each one.

What will happen?

Approve.

Approve.

Approve.

Approve.

Approve.

The human has become a button.

That is not meaningful oversight.

It is organizational theatre.

IBM made this point explicitly in June 2026: “human in the loop” is not by itself a governance strategy if the person cannot meaningfully evaluate the AI’s work, understand the context or challenge the recommendation. [2]

This deserves much more attention.

The existence of a human approval step does not automatically create safety.

The human needs:

information,

time,

authority,

expertise,

and a reason to disagree.

Otherwise the human is not supervising AI.

The human is providing legal decoration.

Meaningful Human Oversight Has Five Requirements

If a person is responsible for reviewing AI work, ask:

1. Can They Understand the Decision?

What happened?

What information mattered?

What policy applied?

What uncertainty exists?

2. Do They Have Enough Time?

If the workflow requires approving hundreds of recommendations continuously, meaningful review may be impossible.

3. Do They Have the Expertise?

An unqualified reviewer does not create strong oversight simply because they are human.

4. Can They Override the AI?

There must be actual authority.

Not merely a button labelled “approve.”

5. Are They Accountable for the Decision?

If nobody owns the final outcome, oversight becomes procedural rather than real.

That is why human-in-the-loop should be designed around decision rights, not checkbox governance.

Mode 4 — AI Autonomous Within Limits

Now we reach the territory that makes executives nervous.

Should an AI agent ever act without asking a human?

Yes.

Companies already allow software to do this constantly.

Fraud systems block transactions.

Algorithms route traffic.

Software allocates workloads.

Trading systems execute instructions.

Cybersecurity systems respond automatically.

The interesting question isn’t whether automation can act.

It is:

Under what boundaries?

Imagine a customer requests a refund.

The agent may be allowed to automatically approve when:

amount < AED 250,

purchase is verified,

policy conditions are satisfied,

no fraud indicators exist,

customer history is normal,

confidence exceeds a defined threshold.

If one condition fails:

escalate.

Now the agent has autonomy.

But it does not have unlimited authority.

This is likely how a large share of agentic work will develop.

Not:

autonomous or not autonomous.

But:

autonomous within explicit boundaries.

Think of Autonomy as a Budget

We give people spending limits.

Why not agents?

Junior manager:

AED 5,000.

Director:

AED 50,000.

CFO:

much higher.

Agents could have equivalent operating boundaries.

For example:

Agent can issue refund:

≤ AED 250.

Agent can offer discount:

≤ 5%.

Agent can reschedule delivery:

yes.

Agent can cancel strategic account contract:

absolutely not.

Agent can generate purchase order:

yes.

Agent can release payment:

only under defined conditions.

This makes autonomy much easier to discuss.

It isn’t philosophical.

It’s governance.

Autonomy Should Follow Risk

A useful principle:

Low Risk + High Reversibility → More Autonomy

High Risk + Low Reversibility → More Human Control

Consider four actions.

Update CRM contact details

Low risk.

Reversible.

High autonomy possible.

Send routine payment reminder

Relatively low risk.

Usually reversible/manageable.

High autonomy may be appropriate.

Offer AED 100 goodwill credit

Moderate but bounded.

Autonomy within thresholds.

Terminate an employee

High consequence.

Difficult to reverse.

Strong human ownership.

This feels obvious when written down.

The problem is many companies are deploying AI without formally writing it down.

The Human-AI Decision Matrix

For every task or decision, score four things.

1. Consequence

If this goes wrong, what happens?

Low → catastrophic.

2. Reversibility

Can we undo the action?

Easy → impossible.

3. Ambiguity

Does the decision follow clear patterns or require deep contextual judgment?

Low → high.

4. Human Value

Does a human interaction itself create:

trust,

relationship,

accountability,

leadership,

or customer value?

Low → high.

Now allocate the work.

This framework is not regulation.

It is a management tool.

And every company should adapt it to its industry.

Add a Fifth Dimension: Confidence

Suppose an AI system normally classifies requests with high reliability.

Why should every request receive identical treatment?

Imagine:

Confidence 99%.

Routine case.

Low risk.

Automate.

Confidence 82%.

Send to human review.

Confidence 51%.

Stop.

Request more information.

This gives us another principle:

Human involvement should increase as machine confidence falls.

Again, not universally.

Some decisions require human review regardless of confidence.

But for many operational workflows, confidence-based escalation is a much smarter design than forcing human involvement into every case.

Humans Should Manage Exceptions

This may become one of the biggest changes in white-collar work.

Today, humans frequently handle:

normal,

normal,

normal,

normal,

normal,

exception,

normal,

normal.

AI-native workflow:

AI handles:

normal,

normal,

normal,

normal,

normal.

Human receives:

exception.

This changes the human role.

People become less responsible for routine execution.

More responsible for:

investigation,

interpretation,

intervention,

and improvement.

Deloitte argues that as agents take on routine work, human roles are likely to place greater emphasis on judgment, investigation and intervention, particularly in edge cases where operational context is required to recognize when agents are wrong. [3]

That sounds attractive.

But it creates a new problem.

If Humans Only See Exceptions, How Will They Learn the Normal Work?

Imagine becoming an accountant who never processes a normal invoice.

You only see the bizarre ones.

Or becoming a lawyer whose first assignments involve only cases the AI could not understand.

Or becoming a junior analyst who never performs foundational research because AI does it.

How do you develop expertise?

For decades, junior professionals learned partly by doing routine work.

The spreadsheet.

The research.

The first draft.

The basic analysis.

The simple customer interaction.

Much of that work is exactly where AI performs well.

This creates a serious talent-development problem.

The company after AI must therefore ask two different questions:

What is the most efficient way to perform this work today?

and:

How will humans develop the expertise we need tomorrow?

Those answers may not always be the same.

Sometimes Humans Should Do Work AI Could Do

Microsoft’s 2026 research found something interesting among its most advanced AI users.

They were more likely than less advanced users to intentionally perform some work without AI to keep their skills sharp. [1]

That is important.

The mature approach to AI isn’t:

“Use AI every possible time.”

It may sometimes be:

“Don’t use AI here because I need to maintain this capability.”

Think about pilots.

They use automation heavily.

They still train for manual control.

Doctors use technology.

They still need clinical expertise.

A CFO may use AI-generated analysis.

They still need the ability to recognize financial nonsense.

Human capability becomes part of organizational resilience.

If employees lose the ability to understand the work, who catches the agent when it is wrong?

The Skills Problem Is Bigger Than Prompting

A lot of AI training focuses on:

how to prompt,

which tools to use,

how to automate.

Useful.

But the more work AI performs, the more valuable another category becomes:

AI supervisory skills.

Can the employee:

define a good objective?

judge AI output?

detect weak assumptions?

identify missing context?

understand uncertainty?

challenge the model?

design escalation?

recognize hallucination?

know when not to automate?

evaluate consequences?

Those are management skills.

Even for employees who never manage another human.

This is why the future workforce conversation isn’t simply:

AI skills.

It is:

skills for working with non-human capability.

Humans Need to Become Better at Asking “Why?”

AI is very good at producing answers.

That changes the value of questions.

If an employee’s entire value was knowing information others didn’t know, AI creates pressure.

If their value is knowing:

which question matters,

which assumption is false,

what the customer isn’t saying,

which trade-off leadership is avoiding,

where the risk actually sits,

then their value may increase.

When answers become cheaper, framing becomes more valuable.

This is one reason the strongest future employees may spend less time producing and more time:

directing,

evaluating,

challenging,

deciding.

Not because production disappears.

Because AI takes a larger share of it.

AI Is Already Changing What “Expert” Means

Historically expertise meant:

I know things.

Increasingly it may mean:

I know how to judge things.

AI can provide information.

Generate possibilities.

Explain technical concepts.

Create analysis.

That increases access to expert-like output.

But experts still possess something difficult to replicate:

pattern recognition built from experience.

They know when the correct-looking answer feels wrong.

That is valuable.

And organizations need to be careful not to remove the very experiences that create that judgment.

What Should AI Own?

So far we have spent a lot of time protecting human work.

Let’s reverse it.

There is plenty of work humans should stop doing.

Humans should probably do less:

copying information,

searching across systems,

formatting,

routine classification,

routine scheduling,

status tracking,

basic reconciliation,

repetitive reporting,

standard follow-up,

document extraction,

information routing,

repeatedly answering identical internal questions.

Why?

Not because those activities have no value.

Because the human brain is expensive.

Use it where it matters.

Udjat’s work on AI for Business in Dubai already makes this practical distinction in marketing: AI should take more repetitive and analytical work while people focus on strategy, creativity and customer understanding.

The same logic should apply across the company.

Do Not Use Humans as APIs

Article 05 used this phrase.

It’s worth repeating.

If an employee spends the day:

opening one system,

reading information,

copying it,

opening another,

typing it,

sending an email,

checking status,

and repeating,

the company is using a human as integration middleware.

That is not dignified work design.

It isn’t particularly good economics either.

Brightery’s work around automating company operations addresses exactly this kind of connected workflow opportunity.

The human should not be the glue simply because your software isn’t connected.

Give the coordination to systems.

Give the person back the decision.

What Should AI Not Own?

There are several categories where companies should be particularly cautious.

Decisions About Human Dignity

Hiring.

Firing.

Promotion.

Disciplinary action.

Compensation.

AI can inform.

Human accountability should remain strong.

Irreversible High-Value Actions

Large financial transfers.

Critical infrastructure actions.

Contract termination.

Major customer commitments.

Autonomy should be tightly bounded.

Decisions Where Context Lives Outside the Data

Not everything important is recorded.

Relationships.

Politics.

History.

Promises.

Culture.

Intent.

AI only sees what it can access.

Humans often know what was never entered into the CRM.

Moments Where Human Presence Is Part of the Product

A luxury customer.

A grieving family.

A difficult employee conversation.

A strategic negotiation.

A major medical conversation.

Efficiency may not be the primary value.

Decisions Without Clear Accountability

If leadership cannot answer:

“Who is responsible if the AI gets this wrong?”

do not increase autonomy yet.

That’s a governance failure disguised as innovation.

Accountability Cannot Be Automated Away

This is one of the most important principles in the series.

Suppose an AI agent rejects a customer request.

Customer complains.

Who is accountable?

The model provider?

Software vendor?

Developer?

Department manager?

Employee supervising the agent?

Chief AI Officer?

CEO?

You cannot answer:

“The AI decided.”

An organization can delegate execution.

It cannot delegate responsibility into a machine and pretend nobody owns the result.

The UAE’s published AI policy principles emphasize exactly this idea: accountability needs to be established for AI outcomes, while human values, safety, transparency and explainability remain central to responsible deployment. [4]

Digital Dubai’s AI principles similarly state that accountability does not sit with the AI system itself and that humans should retain the ability to make final decisions in consequential matters. [5]

That principle will become increasingly important as agents begin to act.

The UAE Context Matters

The UAE is trying to move quickly on AI while simultaneously building a responsible AI environment.

That combination matters.

The conversation should not become:

innovation versus governance.

Governance is part of scaling innovation.

The UAE’s AI policy framework emphasizes principles including:

fairness,

accountability,

transparency,

explainability,

robustness,

safety,

human-centered values,

and privacy. [4]

For UAE companies, that means the human-AI workforce should not be designed purely around productivity.

It also needs to consider:

who is affected,

which information is used,

who can challenge a decision,

how actions can be explained,

where accountability sits,

and whether appropriate human control remains.

That is not simply a compliance exercise.

It’s operating-model design.

AI Changes Management Before It Replaces Jobs

There is another consequence executives should pay attention to.

If AI performs more of the execution, managers become responsible for deciding:

what humans do,

what AI does,

where agents operate,

where approval is required,

how quality is measured,

when exceptions escalate.

That means the manager’s job changes.

The manager stops only allocating work to people.

They begin allocating work across:

people + agents + automation + software.

Microsoft’s 2026 research explicitly frames this as a leadership responsibility: leaders increasingly need to decide what humans and AI do and redesign the system of work around that allocation. [1]

Deloitte goes further, arguing that organizations need to rethink decision rights, accountability and operating models as digital agents begin functioning alongside human workers. [3]

That is not an IT decision.

It’s management.

Imagine the Future Team Meeting

Today’s manager asks:

Ahmed, where are we with the report?

Sara, did you contact the customer?

Omar, can you update the CRM?

Tomorrow’s manager may ask:

Why did the sales agent escalate these 14 opportunities?

Why is the collections agent’s exception rate rising?

Which policy caused the support agent to reject these cases?

Should the procurement agent’s approval threshold increase?

What work are employees still doing manually that the agents should own?

Who is checking quality?

The manager becomes part:

people manager,

process designer,

risk manager,

and agent supervisor.

That is a very different job.

We will return to it later in the series.

The Human-AI Work Allocation Framework

For leadership teams, I would evaluate work in this order.

Step 1 — Define the Outcome

What needs to happen?

Not:

“Write the report.”

But:

“Leadership understands which operating exceptions require action.”

Not:

“Answer support tickets.”

But:

“Customer problems are resolved accurately and quickly.”

Start with outcome.

Step 2 — Break the Workflow Into Decisions and Actions

For example, customer refund:

identify customer,

retrieve transaction,

understand request,

check eligibility,

evaluate exception,

calculate refund,

approve,

execute payment,

communicate,

record.

Do not allocate the whole workflow to “human” or “AI.”

Allocate each component.

Step 3 — Score Each Component

Ask:

How repetitive?

How ambiguous?

How consequential?

How reversible?

How measurable?

How valuable is human interaction?

How good is the available data?

Step 4 — Assign One of Four Modes

HUMAN

Human owns execution and decision.

HUMAN + AI

AI assists.

Human remains deeply involved.

AI + HUMAN APPROVAL

AI performs most work.

Human approves consequential step.

AGENT

AI executes independently within defined limits.

Step 5 — Define Escalation

Even autonomous workflows need exceptions.

What triggers human involvement?

Low confidence?

High transaction value?

Policy exception?

Unusual customer history?

Potential fraud?

Sensitive language?

Regulatory impact?

Make this explicit.

Step 6 — Assign Accountability

Write someone’s name.

Not:

“Operations.”

A person.

Who owns:

performance,

mistakes,

controls,

and improvement?

Example: Sales

Let’s apply it.

Research Account

AI

High information workload.

Low downside.

Human reviews what matters.

Identify Possible Need

Human + AI

AI can detect signals.

Salesperson adds relationship context.

Draft First Outreach

Depending on the business:

AI + Human Approval

or

Agent within approved campaigns.

Discovery Conversation

Usually:

Human + AI support

because trust and ambiguity matter.

Standard Follow-Up

Potentially:

Agent

within defined communication rules.

Negotiate Major Commercial Terms

Human

with AI analytical support.

Update CRM

Why is a human still doing this?

Agent.

This is what job redesign actually looks like.

Not:

“Sales will use AI.”

Too vague.

Example: Finance

Extract Invoice Data

AI / automation

Match Against Purchase Order

Often deterministic software + AI for exceptions.

Identify Anomaly

AI

Investigate Unusual Situation

Human + AI

Approve Routine Low-Value Payment

Potentially automated within controls.

Approve Major Unusual Payment

Human

Create Monthly Management Commentary

AI first draft + human judgment

Decide Capital Allocation

Human + AI

Different work.

Different decision rights.

Example: HR

Answer Routine Policy Question

Agent

Process Standard Leave Request

Potentially Agent within rules

Analyze Employee Survey Themes

AI + Human

Draft Job Description

AI + Human

Evaluate Candidate Evidence

Potentially AI support, with careful governance.

Final Hiring Decision

Human

Performance Coaching

Human + AI preparation

Termination Decision

Human

These distinctions matter because AI touches people directly.

The objective should not be to remove the human from every employee process.

It should be to remove unnecessary administration while protecting meaningful human responsibility.

Example: Marketing

Udjat already operates in an area where the split is becoming visible.

AI can increasingly own:

segmentation,

data analysis,

campaign variations,

routine optimization,

lead scoring,

report generation.

Humans should concentrate more heavily on:

brand strategy,

positioning,

customer understanding,

creative judgment,

commercial priorities,

and deciding what the business should say in the first place.

Udjat’s AI Marketing Automation UAE insight explores how intelligent automation can manage segmentation, lead journeys, personalization and repetitive execution.

But marketing cannot simply become a machine producing infinite content.

Someone must decide:

What deserves to be said?

That is still a human question.

The Danger of AI Abundance

Suppose AI makes:

reports,

analysis,

emails,

images,

videos,

code,

presentations

almost free to create.

What happens?

We may produce far more of all of them.

But someone still needs to:

read,

watch,

review,

approve,

choose,

decide.

AI reduces the cost of production.

It may increase the burden of attention.

This makes human judgment more important.

When 20 options cost almost nothing to generate, the valuable work becomes:

choosing.

When 50 analyses can be produced instantly, the valuable work becomes:

knowing which one matters.

When everyone can create a presentation, the valuable work becomes:

deciding whether there should be a presentation.

This is another reason the company after AI may value judgment more—not less.

Human Attention Becomes the Scarce Resource

AI can generate almost unlimited output.

Humans have:

24 hours.

Management has limited cognitive capacity.

Customers have limited patience.

Employees have limited attention.

So organizations need to stop optimizing only for:

How much can AI produce?

And begin optimizing for:

What deserves human attention?

This is one of the deepest operating-model changes AI creates.

The scarce resource shifts.

From:

production capacity.

Toward:

judgment and attention.

The Human Should Not Become the Company’s Error Department

There is another design trap.

AI handles everything easy.

Humans receive:

angry customer,

strange invoice,

broken workflow,

unusual contract,

ambiguous policy,

agent failure,

security concern.

The human’s entire job becomes:

things going wrong.

That has psychological and operational consequences.

If people interact only with exceptions, work becomes more cognitively demanding.

Roles need to be redesigned intentionally.

Employees may need:

better tools,

fewer cases,

more training,

higher authority,

different performance metrics.

You cannot remove routine work and assume the old job description remains valid.

Measure Human-AI Teams Differently

Traditional productivity metrics may stop making sense.

Imagine a customer-service employee previously handled:

30 cases per day.

With agents, routine cases disappear.

Now the employee handles:

8 cases.

Have they become less productive?

No.

Those eight may be the most complex cases in the entire system.

If management continues measuring:

cases handled per employee

the employee looks worse.

The KPI needs to change.

Perhaps:

complex-case resolution,

customer retention,

exception resolution quality,

decision accuracy,

escalation turnaround.

As work changes, measurement must change with it.

Otherwise the organization will punish employees for adopting the operating model leadership asked them to use.

Reward Outcomes, Not Manual Effort

This is another difficult cultural shift.

Imagine Employee A performs the work manually in eight hours.

Employee B uses AI and finishes in two hours.

What does the company do?

Often:

give Employee B six additional hours of work.

The employee learns something:

Don’t tell anyone you saved time.

That is a terrible incentive system.

If companies want employees to redesign work, they need to reward:

outcomes,

improvement,

automation,

knowledge sharing,

and better systems.

Not simply visible busyness.

Microsoft’s 2026 research found only 13% of surveyed AI users said they were rewarded for reinventing work with AI even when results were uncertain. [1]

That’s a serious organizational problem.

Employees may understand what AI makes possible.

The incentive system still asks them to behave as though nothing changed.

The CEO Needs an AI Decision-Rights Map

I believe every serious AI transformation program will eventually need one.

List important decisions.

Then define:

Who recommends?

Who decides?

Can AI act?

What threshold applies?

What requires human review?

Who owns the result?

For example:

This makes AI governance concrete.

Not:

“Use AI responsibly.”

But:

“AI may do X up to Y under conditions Z.”

That is much more useful.

The Company After AI Needs New Job Descriptions

Today’s job description says:

prepare reports,

manage customer requests,

research accounts,

update CRM,

coordinate meetings,

analyze information.

Tomorrow many of those activities may belong partially or completely to agents.

So perhaps job descriptions need to move from:

Tasks

toward:

Outcomes.

Not:

“Prepare weekly sales report.”

But:

“Ensure management understands risks and opportunities in the sales pipeline.”

Not:

“Respond to customer inquiries.”

But:

“Own resolution of high-value and exceptional customer issues.”

Not:

“Prepare proposals.”

But:

“Design commercially sound solutions that convert qualified opportunities.”

This makes jobs more resilient to changing tools.

The outcome remains.

The method evolves.

A Job Should Not Be Defined by What the Software Cannot Do

This is the same principle from earlier, applied to people.

Don’t design the agent first and then give humans the leftovers.

Design both together.

Ask:

What should the human become exceptional at?

What should the agent become exceptional at?

Where should they collaborate?

Where should responsibility remain unmistakably human?

Deloitte’s 2026 Human Capital Trends research found that only 6% of surveyed leaders said they were making progress designing human-AI interactions, even though work redesign is increasingly critical to capturing AI ROI. [6]

That gap may become one of the biggest reasons some companies benefit significantly more from AI than others.

A Simple Rule for Leaders

When deciding whether AI or a human should own work, ask:

Is the value in performing the task—or in exercising judgment over the outcome?

If the value is primarily execution:

automate aggressively where safe.

If the value is judgment:

augment the human.

If both matter:

design collaboration.

If the consequence is high:

retain clear human accountability.

That is a much better framework than:

“Can AI do this?”

The Future Is Not Humans Versus AI

That framing will look increasingly outdated.

The more interesting competition may be:

Company A

Humans doing everything manually.

versus:

Company B

Humans + AI assistants.

versus:

Company C

Humans + agents working inside redesigned workflows.

The company doesn’t win because it has fewer humans.

It wins because it allocates:

machine capability

and

human capability

more intelligently.

That is the real opportunity.

The UAE Can Design This Early

Many UAE companies are still building or modernizing their operating models.

That creates an advantage.

They do not have to spend decades defining every job around manual information work before redesigning it.

Udjat’s Digital Transformation Agency in Dubai insight already argues that transformation needs to combine technology with process optimization, leadership and measurable business goals.

Human-AI work design is the next extension of that principle.

The question for a growing UAE company should not be:

“How many employees will AI replace?”

It should be:

“How should we design this company now that we have access to both human and artificial intelligence?”

That is a much more ambitious question.

The Human-AI Allocation Checklist

Before assigning any workflow to AI, answer these questions.

Business

What outcome are we trying to create?

What does failure cost?

Judgment

How ambiguous is the work?

Does context matter?

Does expertise matter?

Consequence

Who is affected if the decision is wrong?

How severe could the outcome be?

Reversibility

Can the action be undone?

How quickly?

Human Value

Does trust, empathy, leadership or relationship matter?

AI Capability

Can the system perform the work reliably?

What are its known limitations?

Data

Does the AI have the information required to make the decision?

Oversight

What triggers human intervention?

Does the reviewer have enough information and authority?

Accountability

Whose name sits next to the outcome?

Learning

If AI performs the routine work, how will humans develop and maintain expertise?

This last question should never be forgotten.

What Should Humans Do?

Not everything.

That’s the good news.

People should spend less time being:

data movers,

report generators,

system connectors,

routine classifiers,

follow-up machines.

And more time being:

decision makers,

relationship builders,

problem framers,

leaders,

designers,

negotiators,

investigators,

teachers,

and accountable owners.

What Should AI Do?

More than answer questions.

AI should increasingly handle work that is:

repeatable,

information-heavy,

measurable,

bounded,

and machine-friendly.

It should:

search,

organize,

monitor,

classify,

generate,

compare,

coordinate,

execute,

and escalate.

But its authority should be designed.

Not assumed.

The Answer Is Not a Percentage

People love asking:

“What percentage of this job can AI automate?”

30%?

50%?

80%?

I think that is often the wrong unit.

Jobs are bundles of activities.

Some should disappear.

Some should become AI-assisted.

Some should be autonomous.

Some should become more human.

The future job may not be:

“60% of the old job.”

It may be an entirely different job.

And that is why the AI conversation eventually becomes an organization-design conversation.

The Company After AI Needs Better Humans, Not Just Better AI

As AI becomes more capable, human mediocrity does not necessarily become irrelevant.

It may become more dangerous.

Someone still needs to:

set the objective,

judge quality,

recognize nonsense,

understand consequences,

build trust,

accept responsibility.

If AI gives every employee more capability, the quality of the person’s judgment becomes a multiplier.

Good judgment + powerful AI can produce extraordinary results.

Poor judgment + powerful AI can produce extraordinary mistakes.

Technology increases leverage.

It does not guarantee wisdom.

The Most Important Word Is “Own”

Ask this question for every important workflow:

Who owns the outcome?

Not:

Who clicked the button?

Who wrote the prompt?

Who built the agent?

Who supplied the model?

Who approved the software?

Who owns the result?

If nobody can answer clearly, the workflow is not ready for greater autonomy.

Because in the company after AI, execution may increasingly belong to machines.

Accountability still belongs to people.

And that may be the line that matters most.

What’s Next in The Company After AI

07 — AI Saved 10 Hours. Where Did the Money Go?

Employees say AI saves time.

Vendors report productivity gains.

Managers see work being produced faster.

Then the CFO asks:

“Where is the financial impact?”

That question is becoming increasingly important.

Because:

time saved

is not automatically:

money saved.

In the next article, we’ll examine the AI productivity paradox:

where recovered hours actually go,

why individual productivity does not always become organizational productivity,

how AI can accidentally create more work,

and how CEOs and CFOs should measure AI value differently.

Related Udjat Insights

The human-AI operating model connects naturally with Udjat’s existing work around transformation, automation and responsible AI-enabled marketing:

• Digital Transformation Agency in Dubai: Why Smart Businesses Choose Strategy Before Technology

• How Can AI Change Marketing for Business in Dubai?

• AI Marketing Automation UAE

Series Internal Links

As the previous The Company After AI articles receive their permanent Udjat URLs, connect this article contextually to:

• 01 — Your Company Doesn’t Need an AI Strategy

• 02 — Stop Asking “Where Can We Use AI?”

• 04 — Don’t Automate a Bad Process

• 05 — Your First AI Agent Shouldn’t Be a Chatbot

The strongest connections are Article 04 for work allocation and Article 05 for autonomy, agent permissions and escalation.

Related Brightery Insights

For the technology and workflow implementation behind human-AI teams:

• Automate Company Operations

• AI HR Agent

• AI-Empowered Software: Redefining Smart Business Solutions

• The Best AI Transformation Agency

Sources

[1] Microsoft — 2026 Work Trend Index Annual Report, “Agents, Human Agency, and the Opportunity for Every Organization,” May 5, 2026. Microsoft analyzed Microsoft 365 usage signals and surveyed 20,000 AI-using workers across 10 markets. The research found 49% of analyzed Copilot conversations supported cognitive work; 66% of surveyed AI users said AI allowed more time for higher-value work. Advanced users were also more likely to deliberately decide whether work should be performed by a human or AI and to perform some work without AI to preserve their own skills.

[2] IBM — “Why ‘Human in the Loop’ Alone Is Not a Governance Strategy,” June 17, 2026. IBM argues that merely placing a person at the end of an AI workflow does not constitute meaningful governance. Effective human oversight requires people to possess the context, expertise and authority necessary to evaluate and challenge AI decisions rather than simply rubber-stamping recommendations.

[3] Deloitte — “Rethinking Operating Models for Humans With Agents,” April 2, 2026. Deloitte reports that 84% of surveyed companies had not redesigned jobs around AI and argues that agentic systems create new questions around decision rights, risk, liability, quality and accountability. It suggests human roles will increasingly emphasize judgment, investigation, intervention, oversight and responsibility for edge cases.

[4] UAE Office for Artificial Intelligence — UAE Guiding Principles on AI Policy. The UAE framework emphasizes fairness, accountability, transparency, explainability, robustness, safety, human-centered values and privacy preservation, including the need to establish responsibility for AI impacts.

[5] Digital Dubai — AI Principles and Ethics. Digital Dubai’s principles state that responsibility for AI outcomes does not reside in the AI system itself and emphasize that AI systems should remain safe, controllable and aligned with human values, with people retaining final decision-making capability in consequential matters.

[6] Deloitte — 2026 Global Human Capital Trends, March 4, 2026. Deloitte reports that only 6% of surveyed leaders say they are making progress designing effective human-AI interactions, despite work redesign becoming increasingly important to realizing AI value. The report also found that 65% of surveyed organizations believe their culture needs to change significantly because of AI.

[7] Deloitte — “Rewiring the Enterprise Operating Model for AI Scale,” July 2026. Deloitte argues that scaled AI requires organizations to redesign decision rights, workforce structures, accountability and coordination as AI becomes embedded throughout enterprise workflows.

[8] IBM — “From AI Governance to AI Assurance: What We Shared at Think 2026,” June 16, 2026. IBM describes enterprise AI as an interconnected environment of models, agents, workflows, data, controls and decisions, arguing for continuous visibility, enforceable controls and explicit ownership as AI systems gain greater operational authority.