How AI Can Augment a People-Centered Workforce: A Practical Framework for Redesigning Jobs, Developing Skills, Building Trust, and Combining Human Judgment with Intelligent Automation

1. Purpose: Begin with business, customer, and employee outcomes.

2. Work design: Analyze jobs at the task level and assign activities appropriately.

3. Skills: Build AI literacy, technical capability, critical thinking, and judgment.

4. Trust: Be transparent about how AI is used and how decisions are made.

5. Governance: Establish rules for privacy, accuracy, bias, security, oversight, and accountability.

6. Measurement: Track quality, employee experience, adoption, risk, productivity, and business value.

The World Economic Forum’s Future of Jobs Report 2025 found that AI and information-processing technologies are among the most consequential forces expected to reshape employment and skills through 2030. The report also emphasizes that technological skills will grow in importance alongside human capabilities such as analytical thinking, resilience, leadership, and collaboration. OECD research involving employers and workers has found generally positive views about AI’s effects on performance and working conditions, while also identifying concerns about job quality, skills, privacy, work intensity, and worker participation.

The most important principle is:

AI should not be introduced only to make the existing organization faster. It should be used to redesign work so that people, technology, and business processes operate better together.

1. What Is a People-Centered AI Workforce?

A people-centered AI workforce is an operating model in which artificial intelligence is designed, implemented, and governed around human needs, capabilities, rights, responsibilities, and experiences. It does not assume that every task currently performed by a person should remain unchanged. It also does not assume that every technically automatable task should immediately be transferred to a machine.

Instead, the organization evaluates how work should be divided among:

• Employees
• Managers
• Technical specialists
• External partners
• Traditional automation
• AI copilots
• Autonomous agents

The objective is to improve the complete system of work.

That may involve:

• Reducing repetitive administrative work
• Improving access to knowledge
• Supporting faster decisions
• Increasing accuracy
• Creating more personalized customer service
• Improving employee learning
• Expanding organizational capacity
• Strengthening accessibility
• Reducing workplace frustration

EY’s framework begins with the principle that people remain the ones who imagine, build, use, and experience changes in work. It argues that organizations need to combine operational benefits with investment in employee skills, career development, culture, and experience.

2. Why People-Centered Does Not Mean Technology-Resistant

People-centered design is sometimes misunderstood as an effort to protect every existing job, process, and responsibility from technological change. That is not the objective. Some work should be automated. Some roles will shrink. Some responsibilities will move. Some skills will become less valuable. New roles and capabilities will emerge.

A people-centered strategy accepts these realities while insisting that the transformation should be:

• Purposeful
• Transparent
• Fair
• Secure
• Governed
• Supported by learning
• Connected to clear business outcomes

The alternative is technology-centered implementation.

A technology-centered initiative often begins with a model or software product and asks:

Where can we deploy this?

A people-centered initiative begins with a problem and asks:

What are employees or customers trying to accomplish, what prevents them from succeeding, and what combination of people and technology would improve the outcome? This difference affects everything that follows.

3. Technology Adoption Without Work Redesign Usually Underperforms

Organizations frequently add new technology to old processes. An employee completes a long approval form. AI helps fill in the form faster. The approval process still contains unnecessary stages. A customer waits for service. AI generates a response. The employee still needs to copy information among disconnected systems. A manager receives more AI-generated reports. The organization still lacks clear decision rights. EY warns that combining new technology with outdated processes can produce an expensive version of the old process rather than genuine transformation.

The correct sequence is:

1. Define the required outcome.

2. Understand the current workflow.

3. Identify unnecessary activities.

4. Decide which tasks should be automated or augmented.

5. Redesign roles and responsibilities.

6. Introduce the technology.

7. Measure results and improve the process.

Technology should support redesign, not replace it.

4. Augmentation Versus Automation

The distinction between augmentation and automation is central to a responsible AI strategy.

4.1 Automation

Automation transfers responsibility for executing a task to a system.

Examples include:

• Routing an invoice
• Scheduling an appointment
• Generating a recurring report
• Categorizing a standard request
• Monitoring defined system conditions
• Updating approved records

Automation works best when:

• Rules are clear
• Inputs are reliable
• Outputs are measurable
• Exceptions can be identified
• Errors are reversible
• Risk is limited

4.2 Augmentation

Augmentation helps a person perform a task while the person retains judgment or accountability.

Examples include:

• Suggesting possible diagnoses to a clinician
• Drafting a customer response
• Summarizing legal documents
• Identifying unusual financial activity
• Recommending code improvements
• Creating a first version of a presentation
• Finding relevant organizational knowledge

EY describes generative AI as potentially producing first drafts of reports and analyses so employees can concentrate on assessment, refinement, and subsequent action.

Augmentation is particularly valuable when:

• Human context matters
• The problem is ambiguous
• The output requires verification
• Relationships matter
• Accountability cannot be delegated
• Errors could have significant consequences

4.3 Autonomous agent execution

A third category is emerging between traditional automation and full human control.

An AI agent may execute a sequence of actions, such as:

• Investigating a routine support case
• Updating a customer record
• Monitoring infrastructure
• Preparing a procurement request
• Coordinating a workflow

The agent may operate independently within defined boundaries.

This requires explicit decisions about:

• Authorized actions
• Data access
• Spending limits
• Escalation
• Human approval
• Audit trails
• Emergency shutdown
• Accountability

A people-centered strategy does not assume that autonomous activity is inappropriate. It requires stronger governance as agent authority increases.

5. Begin With the Employee and Customer Outcome

AI implementation should start with a clearly defined outcome.

Weak objectives include:

• Deploy generative AI
• Increase AI usage
• Introduce copilots
• Automate more work
• Reduce headcount

Stronger objectives include:

• Reduce the time nurses spend searching for patient information.
• Help service representatives resolve customer problems during the first interaction.
• Reduce the administrative burden on teachers.
• Help engineers identify software defects earlier.
• Give employees faster access to approved organizational knowledge.
• Improve the consistency of regulatory reporting.
• Reduce repetitive manual data entry.

The difference is important. The first group measures technology activity. The second group describes human and business value.

6. Analyze Jobs at the Task Level

A job is not one indivisible activity. It is a collection of tasks.

A recruiter may:

• Draft job advertisements
• Search for candidates
• Review applications
• Schedule interviews
• Advise hiring managers
• Evaluate candidate fit
• Negotiate offers
• Maintain relationships

AI might help draft job descriptions, organize candidate information, schedule interviews, and identify search terms.

Human recruiters may retain responsibility for:

• Assessing context
• Evaluating motivation
• Managing relationships
• Challenging biased requirements
• Negotiating
• Making accountable recommendations

A task-level analysis makes the transformation more precise.

For every task, ask:

• How frequently is it performed?
• How much time does it require?
• Is it repetitive?
• Does it rely on stable rules?
• Is the necessary data available?
• How serious would an error be?
• Does it require empathy or judgment?
• Who is accountable?
• Can the result be verified?

7. A Human-AI Work Allocation Framework

Tasks can be assigned to five categories.

7.1 Eliminate

Some tasks should not continue.

Examples include:

• Duplicate reporting
• Unused approvals
• Manual transfers created by poor system integration
• Meetings without a clear purpose

The first productivity improvement may come from stopping work rather than automating it.

7.2 Automate

Transfer appropriate structured and repetitive tasks to software.

7.3 Augment

Use AI to support a person who remains responsible.

7.4 Delegate with supervision

Allow an agent to execute a controlled workflow while a person monitors performance and handles exceptions.

7.5 Preserve as human-led

Keep direct human control when the activity requires strong judgment, trust, accountability, empathy, or ethical responsibility.

This framework avoids two extremes:

• Automating everything technically possible
• Preserving every existing responsibility because change is uncomfortable

8. Use AI to Reduce Low-Value Friction

Employees spend substantial time performing work that is necessary but not inherently valuable.

Examples include:

• Searching for information
• Reformatting documents
• Reentering data
• Preparing routine summaries
• Finding policies
• Scheduling
• Creating status reports
• Navigating several systems

AI can improve the employee experience by reducing these forms of friction. A knowledge assistant may allow an employee to ask a question in natural language and receive information from approved internal sources. A meeting assistant may create notes, decisions, and action items. A finance assistant may explain a variance and identify relevant supporting data. The objective is not only time savings.

Reducing friction can also improve:

• Focus
• Accuracy
• Responsiveness
• Employee satisfaction
• Customer service
• Organizational learning

9. Preserve Meaningful Human Work

Not all efficiency improvements improve the experience of work.

An AI system may accelerate task completion while making the employee feel:

• Constantly monitored
• Less trusted
• More interchangeable
• Responsible for correcting machine errors
• Pressured to handle greater volume
• Excluded from decisions

A people-centered strategy evaluates not only whether AI reduces task time but also whether it improves:

• Autonomy
• Purpose
• Skill use
• Learning
• Workload
• Psychological safety
• Job quality

OECD research indicates that workers and employers often see benefits in AI-related productivity and working conditions, but outcomes depend significantly on implementation, consultation, training, and workplace practices.

10. Productivity Gains Require a Deliberate Allocation Decision

Suppose an AI assistant reduces the time required for a task by 30 percent. The organization must decide what happens to the released capacity.

It might be used to:

• Reduce staffing
• Serve more customers
• Improve quality
• Shorten waiting times
• Increase innovation
• Reduce overtime
• Provide more personalized service
• Allow employees to learn
• Build stronger customer relationships

If leaders do not make this decision explicitly, the default may become increased work volume. Employees may be expected to complete more tasks without seeing any improvement in job quality. This can weaken trust and adoption.

A credible AI strategy should explain how gains will be shared among:

• Customers
• Employees
• Investors
• The organization
• Society

11. Trust Is an Operating Requirement

Employees are less likely to adopt AI when they believe it is:

• A hidden surveillance mechanism
• A preliminary step toward dismissal
• Unreliable
• Unfair
• Unaccountable
• Designed without their input

Trust does not come from communication campaigns alone. It comes from observable organizational behavior.

Employees need clarity about:

• Why AI is being introduced
• Which tasks it will affect
• What data it uses
• How performance will be evaluated
• Whether outputs can be challenged
• Who remains accountable
• What training will be available
• How workforce changes will be managed

Transparency does not require disclosing every technical detail. It requires giving people enough information to understand how the system affects their work and rights.

12. Involve Employees in Work Redesign

Employees often understand the practical workflow better than executives, consultants, or software vendors.

They know:

• Where customers become frustrated
• Which steps are unnecessary
• Where data is unreliable
• Which exceptions occur
• Which shortcuts people use
• Which decisions require experience

Worker participation can improve:

• Use-case selection
• Process design
• Risk identification
• Training
• Adoption
• Trust

A good design process may include:

• Interviews
• Workshops
• Process mapping
• Pilot groups
• Feedback sessions
• Employee councils
• User testing

Participation does not mean every employee receives veto power over change. It means decisions are informed by the people who understand and experience the work.

13. AI Literacy Is Becoming a Foundational Workforce Skill

AI literacy does not mean every employee must become a machine-learning engineer. It means people understand enough to use AI appropriately.

A basic AI-literacy program should cover:

• What the tool can do
• What it cannot do reliably
• How to provide useful instructions
• How to verify outputs
• What data may be entered
• Which decisions require human review
• How bias and errors can arise
• How to report a problem

Different roles require different levels of knowledge.

General employees may need:

• Safe use
• Verification
• Privacy awareness
• Basic prompting
• Escalation procedures

Managers may need:

• Work redesign
• Performance measurement
• Change leadership
• Ethical decision-making
• Human-agent management

Technical professionals may need:

• Model architecture
• Evaluation
• Security
• Data pipelines
• Monitoring
• Agent controls

Executives may need:

• Strategic implications
• Investment economics
• Risk appetite
• Governance
• Workforce consequences

14. Training Must Be Connected to Real Work

EY noted in its 2023 research that AI-related training was not yet a leading priority for many employers or employees, even as adoption expectations increased. It recommended developing a roadmap for training people to use AI in ways that empower them to focus on higher-value tasks. Training is most effective when employees can apply it immediately.

Useful approaches include:

• Role-based learning
• Guided practice
• AI sandboxes
• Peer coaching
• Office hours
• Communities of practice
• Approved prompt libraries
• Real project assignments
• Manager-supported experimentation

A generic course followed by no opportunity to use the tool will produce limited capability.

15. Managers Are Central to AI Adoption

Employees experience organizational change primarily through their immediate manager.

Managers must be able to explain:

• Why the change is happening
• How roles may evolve
• Which tasks are affected
• How performance will be evaluated
• Which training is available
• What remains uncertain

Managers also need to redesign work.

They must decide:

• What should be delegated to AI
• What requires review
• How workload should change
• How team responsibilities should shift
• How junior employees will continue learning

Poorly prepared managers can turn a technically promising implementation into a workforce failure.

16. Redesign Performance Management

Traditional performance systems may reward:

• Hours worked
• Volume
• Speed
• Individual output
• Visible busyness

AI changes the relationship between effort and output. An employee using AI effectively may produce more work in less time. Another employee may spend time carefully verifying a high-risk result.

Performance management should increasingly consider:

• Quality
• Judgment
• Business impact
• Collaboration
• Responsible AI use
• Customer outcomes
• Learning
• Improvement of team systems

Organizations should not reward employees for maximizing AI-generated volume regardless of quality.

17. Protect the Development of Junior Employees

Early-career workers often learn through tasks such as:

• Research
• Drafting
• Documentation
• Basic analysis
• Routine coding
• Customer support
• Data preparation

AI can perform portions of this work. If organizations simply remove junior tasks, they may weaken the future pipeline of experienced professionals.

New development models may include:

• Apprenticeships
• Simulation
• Rotations
• Structured review
• Human-AI pair work
• Customer exposure
• Mentoring
• Supervised agent management

The objective should be to accelerate development rather than eliminate the ladder by which people become experts.

18. AI Can Personalize Employee Learning

AI systems can support learning by:

• Recommending content
• Explaining concepts
• Generating practice exercises
• Simulating conversations
• Providing immediate feedback
• Translating materials
• Adapting difficulty
• Identifying skill gaps

However, AI recommendations should not become unreviewed decisions about employee potential.

Employees should be able to understand and challenge:

• Skill classifications
• Training recommendations
• Career suggestions
• Performance interpretations

Human managers and employees should remain participants in development decisions.

19. AI in Recruitment and Talent Management

AI can assist with:

• Writing job descriptions
• Candidate search
• Interview scheduling
• Onboarding
• Learning recommendations
• Workforce analytics
• Internal mobility

These uses can improve speed and access. They can also introduce significant risk. Historical data may reproduce existing patterns. A system may disadvantage candidates based on proxies for protected characteristics. Automated recommendations may appear objective even when the underlying model is incomplete. High-impact employment decisions require strong oversight, testing, documentation, and legal review.

20. Human Oversight Must Be Specific

“Human in the loop” is often used as a broad assurance.

It is not meaningful unless the organization defines:

• Which human is responsible
• What information the person receives
• Whether the person has authority to override the system
• Whether the person has enough time to review
• What training is required
• How disagreement is recorded
• How repeated errors are escalated

A person who mechanically approves hundreds of AI decisions is not providing meaningful oversight.

21. Accountability Cannot Be Assigned to the Algorithm

AI may recommend or execute an action. The organization remains accountable for deploying and governing the system.

Responsibility should be assigned for:

• Use-case approval
• Data quality
• Model performance
• Security
• Legal compliance
• Employee impact
• Customer impact
• Incident response
• System suspension

NIST’s AI Risk Management Framework is designed to help organizations manage risks to individuals, organizations, and society throughout the design, deployment, and use of AI systems. It emphasizes governance and trustworthy AI characteristics rather than assuming that technical performance alone is sufficient.

22. Manage Accuracy and Reliability

Generative AI can produce confident but incorrect outputs.

Risks include:

• Fabricated information
• Incomplete answers
• Incorrect calculations
• Misleading summaries
• Inconsistent results
• Outdated information

EY specifically identifies accuracy and reliability as issues that organizations should address when implementing generative AI.

Controls may include:

• Approved data sources
• Retrieval from verified knowledge
• Output testing
• Human review
• Confidence thresholds
• Restricted use cases
• Monitoring
• Feedback mechanisms
• Incident reporting

The required control level should increase with the potential harm.

23. Protect Privacy and Confidentiality

Employees may accidentally enter:

• Customer information
• Employee records
• Financial data
• Source code
• Legal documents
• Strategic plans

into public or unapproved AI systems.

Organizations need clear rules covering:

• Approved tools
• Permitted data
• Prohibited data
• Data retention
• Vendor use of inputs
• Model training
• Cross-border transfers
• Access controls
• Logging

Privacy should be built into the implementation rather than handled only through employee warnings.

24. Address Bias and Fairness

AI systems can reproduce or amplify bias through:

• Historical data
• Incomplete datasets
• Proxy variables
• Unequal error rates
• Biased labels
• Poorly designed evaluation

Workforce uses deserve particular scrutiny because they can affect:

• Hiring
• Promotion
• Compensation
• Scheduling
• Performance
• Termination
• Access to development

EY highlights the need to evaluate whether AI systems reflect organizational commitments relating to diversity, equity, inclusion, and culture. Fairness requires more than removing explicitly sensitive attributes.

It may require:

• Impact testing
• Representative data
• Independent review
• Employee appeal
• Ongoing monitoring
• Human accountability

25. Cybersecurity Is Part of Workforce AI

AI systems may connect with:

• Email
• Documents
• HR systems
• Customer databases
• Code repositories
• Financial systems
• Collaboration platforms

This can create risks such as:

• Unauthorized data access
• Prompt injection
• Credential misuse
• Malicious content
• Excessive agent permissions
• Data leakage
• Manipulated outputs

Security controls should include:

• Least-privilege access
• Identity management
• Tool approval
• Data segregation
• Activity logging
• Incident response
• Agent permission limits
• Regular testing

A workforce AI system may appear to be an HR or productivity application while possessing access to highly sensitive enterprise information.

26. Consider Cost, Scale, and Infrastructure

AI implementation costs can include:

• Software subscriptions
• Model usage
• Data preparation
• Integration
• Security
• Monitoring
• Training
• Change management
• Human review
• Vendor support
• Infrastructure

EY recommends evaluating model selection, performance, cost, knowledge infrastructure, partnerships, and the broader talent benefits involved in implementation. A small pilot may appear inexpensive.

Enterprise deployment may require substantial investment in:

• Identity
• Data access
• Governance
• Support
• Reliability
• Evaluation
• Adoption

Organizations should evaluate total cost rather than model price alone.

27. Do Not Confuse Adoption With Value

Common AI adoption metrics include:

• Number of licenses
• Active users
• Prompts submitted
• Documents generated
• Agent tasks completed

These measures can indicate activity. They do not prove value.

Better measures include:

Employee outcomes

• Time saved
• Reduced administrative burden
• Confidence
• Satisfaction
• Learning
• Ability to focus
• Workload

Business outcomes

• Revenue
• Customer satisfaction
• Resolution time
• Quality
• Error reduction
• Speed to market
• Cost per transaction

Risk outcomes

• Incorrect outputs
• Privacy incidents
• Security events
• Bias findings
• Override rates
• Escalations

Adoption quality

• Appropriate use
• Repeat use
• Verification behavior
• Use-case completion
• Manager support

EY recommends measuring workforce confidence, sentiment, and adoption so organizations can adjust their strategy during implementation.

28. Employee Experience Must Be Measured Directly

Leaders should not assume that faster task completion means employees are having a better experience.

Useful questions include:

• Does the tool reduce frustration?
• Does it increase or reduce autonomy?
• Do employees trust the output?
• Does it create more monitoring?
• Does it increase workload?
• Does it help employees serve customers?
• Are people learning?
• Do employees understand how decisions are made?

Employee surveys should be supplemented with:

• Interviews
• Usage analysis
• Workflow observations
• Error reports
• Focus groups
• Support data

29. AI Can Support Accessibility and Inclusion

AI may help employees through:

• Speech-to-text
• Text-to-speech
• Translation
• Summarization
• Alternative formats
• Communication assistance
• Personalized interfaces

These capabilities can improve access for:

• Employees with disabilities
• Non-native speakers
• People working across languages
• Employees with different learning needs

However, accessibility tools should be tested with the people they are intended to support. A system that works well in a demonstration may fail in real working conditions.

30. Do Not Use AI Primarily as Employee Surveillance

AI can analyze:

• Communication
• Activity
• Keystrokes
• Meetings
• Productivity
• Location
• Sentiment

The technical ability to collect data does not justify doing so.

Excessive monitoring can damage:

• Trust
• Autonomy
• Psychological safety
• Morale
• Retention

Workplace analytics should have:

• A defined purpose
• Proportionality
• Privacy safeguards
• Access controls
• Retention limits
• Employee communication
• Human review

The purpose should be to improve systems of work, not create invisible behavioral control.

31. Workforce Transitions Must Be Managed Honestly

AI may reduce demand for some work. A people-centered strategy does not require denying this possibility. It requires handling change responsibly.

Possible actions include:

• Redeployment
• Reskilling
• Reduced hiring
• Natural attrition
• Role redesign
• Voluntary transitions
• Separation support

Employees should not be told that AI will only help them if leadership is already planning significant role reductions. Honest communication is difficult, but misleading communication destroys trust.

32. Build an AI Workforce Governance Model

A practical governance model should include several levels. Executive oversight

Executives define:

• Strategic goals
• Risk appetite
• Investment priorities
• Workforce principles
• Accountability

AI governance function

This group may establish:

• Policies
• Risk classifications
• Approval processes
• Evaluation standards
• Incident procedures

Business and product owners

They remain responsible for:

• Use-case outcomes
• Workflow design
• Adoption
• Human impact
• Operating performance

Technical teams

They manage:

• Models
• Data
• Integration
• Security
• Monitoring
• Reliability

HR and workforce leaders

They manage:

• Skills
• Role redesign
• Employee communication
• Training
• Workforce transitions
• Employment-policy implications

Employees

They provide:

• Feedback
• Operational knowledge
• Error reports
• Improvement ideas
• Responsible use

33. Use Risk-Based Governance

Not every AI use case requires the same level of control. Lower-risk use cases

• Brainstorming
• Formatting
• Drafting low-sensitivity internal text
• Meeting summaries

Moderate-risk use cases

• Customer-service recommendations
• Internal knowledge retrieval
• Financial analysis
• Code generation

High-risk use cases

• Hiring recommendations
• Medical decisions
• Credit decisions
• Safety controls
• Employee discipline
• Legal conclusions
• Autonomous financial transactions

Controls should increase with:

• Impact
• Autonomy
• Data sensitivity
• Difficulty of reversal
• Number of affected people

34. Create an AI Use-Case Review

Before deployment, evaluate:

• Business objective
• Employee objective
• Customer impact
• Data source
• Accuracy requirements
• Bias risk
• Privacy
• Security
• Human oversight
• Cost
• Vendor dependence
• Exit plan

The review should be efficient enough to support innovation but rigorous enough to prevent unmanaged deployment.

35. Pilot With Real Work

A useful pilot should involve:

• A clearly defined workflow
• Realistic data
• Representative employees
• Measurable outcomes
• Risk controls
• Feedback
• Comparison with the current process

A demonstration showing that AI can generate text is not a workforce pilot. A pilot should test whether the complete human-AI system performs better.

36. Define Success Before the Pilot Begins

Possible success criteria include:

• Reduced handling time
• Improved accuracy
• Higher employee satisfaction
• Better customer experience
• Lower rework
• Reduced search time
• Faster onboarding
• Increased accessibility

The organization should also define unacceptable outcomes, such as:

• Increased error rates
• Privacy violations
• Lower trust
• Unmanageable review burden
• Unequal impact
• Excessive cost

37. Scale Only After the Operating Model Works

A pilot may succeed because:

• Enthusiastic employees volunteered.
• Experts supported every step.
• The dataset was small.
• Exceptions were handled manually.
• Costs were temporarily subsidized.

Scaling requires:

• Support
• Training
• Documentation
• Monitoring
• Governance
• Integration
• Cost control
• Manager capability
• Incident response

The organization should not assume pilot economics and performance will automatically continue at enterprise scale.

38. A Practical People-Centered AI Framework

A complete strategy can be organized around eight principles. Principle 1: Purpose before technology Begin with an employee, customer, or business outcome. Principle 2: Redesign before automation Improve the process before applying AI. Principle 3: Tasks before jobs Analyze activities rather than treating occupations as indivisible. Principle 4: Augmentation before replacement Consider whether AI can increase human effectiveness before removing human involvement. Principle 5: Accountability remains human Assign clear ownership for AI-enabled decisions and actions. Principle 6: Skills and participation

Train employees and involve them in redesign. Principle 7: Proportional governance Apply controls according to risk, autonomy, and impact. Principle 8: Measure human and business value Track employee experience, quality, risk, and outcomes, not only adoption.

39. A 12-Month Implementation Roadmap

Months 1 to 2: Establish principles and governance

• Define the people-centered AI philosophy.
• Appoint executive sponsors.
• Establish approved-use policies.
• Define risk categories.
• Identify prohibited uses.

Months 3 to 4: Identify priority workflows

• Map employee pain points.
• Evaluate customer problems.
• Identify repetitive work.
• Assess data readiness.
• Select pilot use cases.

Months 5 to 6: Redesign work

• Break roles into tasks.
• Eliminate unnecessary activities.
• Define human and AI responsibilities.
• Establish review and escalation.
• Design role changes.

Months 7 to 8: Pilot

• Train employees.
• Deploy controlled tools.
• Measure quality and time.
• Collect employee feedback.
• Monitor risk and cost.

Months 9 to 10: Improve the operating model

• Refine workflows.
• Update training.
• Strengthen security.
• Adjust oversight.
• Improve integrations.
• Clarify performance expectations.

Months 11 to 12: Scale selectively

• Expand successful use cases.
• Stop low-value pilots.
• Update workforce plans.
• Build communities of practice.
• Establish quarterly governance reviews.

40. What Small and Midsize Organizations Should Do

Smaller organizations do not need a large AI-governance bureaucracy. They do need clear rules.

A practical approach is:

1. Approve a limited number of tools.

2. Define which data may be entered.

3. Select a few high-value, low-risk use cases.

4. Train employees in verification and privacy.

5. Require human approval for important decisions.

6. Measure real outcomes.

7. Review vendor terms carefully.

Smaller companies should avoid creating many disconnected subscriptions without visibility or governance.

41. What Large Enterprises Should Do

Large enterprises need more structured systems because they have:

• More employees
• More sensitive data
• More jurisdictions
• More vendors
• More use cases
• Greater regulatory exposure

They may require:

• Enterprise AI policy
• Risk classification
• Tool inventory
• Model inventory
• Workforce consultation
• Technical evaluation
• Role-based training
• Regional legal review
• Incident reporting
• Executive governance

The challenge is to maintain control without making responsible adoption impossible.

42. Common Failure: Starting With Headcount Reduction

When AI is introduced primarily as a workforce-reduction target, employees may resist, hide information, or avoid helping redesign the work. Cost reduction may be a legitimate objective.

It should be evaluated alongside:

• Quality
• Growth
• service
• innovation
• resilience
• employee capability

43. Common Failure: Automating the Visible Task but Ignoring the Workflow

An AI system may accelerate one step while the complete process remains slow. Leaders should measure end-to-end performance.

44. Common Failure: Providing Tools Without Training

Employees may:

• Avoid the tool
• Use it incorrectly
• Enter confidential information
• Trust incorrect output
• Create inconsistent practices

Technology access is not capability.

45. Common Failure: Treating Human Review as an Unlimited Resource

AI may generate more content than employees can realistically check. The cost and time of verification must be included in the business case.

46. Common Failure: Ignoring Employee Emotion

Workforce transformation involves uncertainty, status, identity, and fear. Leaders should not treat employee concerns as simple resistance to innovation. Some concerns may reveal genuine design and governance problems.

47. Common Failure: Measuring Only Time Saved

Time savings may be offset by:

• Rework
• Review
• Errors
• Security risks
• Lower trust
• Increased workload

A balanced measurement system is necessary.

48. Common Failure: Scaling Before Trust Is Established

A technically successful system can fail if employees do not understand or trust it. Adoption should be built through participation, transparency, support, and visible accountability.

Key Takeaways

1. A people-centered AI strategy focuses on what people can accomplish with technology, not only what the technology can do.

2. Augmentation and automation are different.

Augmentation supports human work, while automation transfers task execution to a system.

3. AI implementation should begin with employee, customer, and business outcomes.

4. Existing processes should be redesigned before they are automated.

5. Jobs should be analyzed at the task level.

6. Human judgment should remain central in high-impact, ambiguous, ethical, and relationship-based work.

7. Productivity gains should be allocated intentionally rather than automatically becoming increased workload.

8. Employee trust is an operating requirement, not a communication accessory.

9. Workers should participate in redesign because they understand real workflows and exceptions.

10. AI literacy must become a broad workforce capability.

11. Training should be role-specific and connected to real work.

12. Managers play a central role in work redesign, communication, and adoption.

13. Junior career pathways must be redesigned as AI absorbs routine developmental tasks.

14. Human oversight must include authority, time, information, and accountability.

15. Privacy, security, accuracy, and bias should be managed from the beginning.

16. Workforce AI should not become an uncontrolled employee-surveillance system.

17. Adoption metrics do not prove value.

18. Employee experience, quality, business outcomes, and risk should be measured together.

19. Governance should be proportional to the impact and autonomy of each use case.

20. Successful AI transformation redesigns the complete system of people, processes, skills, technology, and accountability.

Frequently Asked Questions

What is a people-centered AI workforce?

It is a workforce model in which AI is designed and governed around human needs, capabilities, rights, responsibilities, and experiences.

What does AI augmentation mean?

AI augmentation means using AI to improve human performance while a person retains important judgment, review, or accountability.

What is the difference between augmentation and automation?

Augmentation helps a person perform the work. Automation transfers execution of the task to a system.

Will AI replace employees?

AI will automate portions of many jobs and may reduce demand for some roles. It will also change existing jobs and create new responsibilities. The outcome depends on the task, industry, technology, economics, and organizational choices.

Which tasks are most suitable for automation?

Tasks are stronger candidates when they are:

• Repetitive
• Structured
• Measurable
• Low risk
• Based on reliable data
• Easy to verify

Which tasks should remain human-led?

Human leadership is especially important for:

• Ethical judgment
• Sensitive relationships
• Leadership
• High-stakes decisions
• Ambiguous strategy
• Legal accountability
• Complex negotiation

What is a human-in-the-loop system?

It is a system in which a person reviews, approves, corrects, or supervises AI output or action. The human role must be clearly defined to be meaningful.

How can AI improve employee experience?

It can reduce repetitive administration, improve information access, support learning, increase accessibility, and help people focus on higher-value work.

Can AI make employee experience worse?

Yes. Poor implementation may increase surveillance, workload, error correction, anxiety, and loss of autonomy.

How should employees be trained?

Training should cover:

• Tool capabilities
• Limitations
• Verification
• Privacy
• Security
• Responsible use
• Escalation

It should be tailored to the employee’s role.

Should employees be involved in AI design?

Yes. Employees can identify practical workflow problems, risks, exceptions, and opportunities that may not be visible to leaders or vendors.

How should AI productivity be measured?

Useful measures include:

• Time saved
• Quality
• Error rate
• Customer outcomes
• Employee experience
• Cost
• Adoption quality
• Risk events

Is the number of AI users a useful metric?

It measures activity but does not prove that AI creates value.

How can organizations build trust?

They should communicate clearly, involve employees, establish accountability, protect data, provide training, and allow errors or decisions to be challenged.

Can AI be used in hiring?

It can assist some activities, but hiring systems require strong oversight because they can introduce bias, privacy risks, and unfair decisions.

Can AI be used for performance management?

It may support analysis, but automated employment decisions can be high risk. Human review, transparency, legal assessment, and appeal mechanisms are important.

What data should employees enter into AI tools?

Only data permitted by the organization’s policy and vendor agreement. Sensitive, personal, confidential, regulated, and proprietary information should not be entered into unapproved systems.

What is responsible AI governance?

It is the set of policies, roles, technical controls, reviews, monitoring, and accountability mechanisms used to manage AI risk.

Who should be accountable for an AI decision?

An identified person, business owner, or governance body should remain accountable. Accountability should not be assigned vaguely to the model or vendor.

How should small companies govern AI?

They should approve tools, define data rules, select low-risk use cases, train employees, require human review for important decisions, and monitor outcomes.

How can companies protect junior career development?

They can create apprenticeships, structured review, simulations, rotations, mentoring, customer exposure, and supervised human-AI work.

Should AI be used to reduce headcount?

AI may change workforce capacity requirements. Decisions should consider strategy, service, quality, fairness, redeployment, reskilling, and long-term capability rather than focusing only on immediate cost.

Conclusion

Artificial intelligence gives organizations an extraordinary opportunity to redesign work. It can help employees find information, create first drafts, analyze data, automate administration, monitor operations, and coordinate complex workflows. These capabilities can increase productivity and organizational capacity. They can also improve the experience of work by reducing frustration and allowing people to focus on customers, decisions, creativity, and relationships. None of these outcomes is automatic. A poorly designed AI implementation can increase surveillance, workload, confusion, risk, and employee anxiety. It can automate unnecessary processes, produce unreliable output, reinforce bias, expose sensitive information, and weaken the development of future talent. The deciding factor is not simply the quality of the model. It is the quality of the organizational design surrounding the model. A people-centered workforce strategy begins with purpose. It identifies what employees, customers, and the business need to accomplish. It examines the complete workflow.

It removes unnecessary work. It assigns tasks according to the strengths of people, automation, copilots, and agents. It gives employees the skills and authority required to use AI responsibly. It establishes clear human accountability. It measures whether quality, trust, employee experience, and business outcomes are actually improving. Most importantly, it recognizes that people are not an obstacle between the organization and automation. People are the source of context, judgment, ethics, relationships, imagination, and responsibility that makes technology valuable.

The defining question is not:

How much human work can AI replace?

It is:

How should we redesign work so that people can create greater value, exercise better judgment, develop stronger capabilities, and experience more meaningful work with the support of intelligent technology?

Relevant Articles and Resources

1. How Artificial Intelligence Can Augment a People-Centered Workforce

EY

https://www.ey.com/en_gl/insights/workforce/how-artificial-intelligence-can-augment-a-people-centered-workforce

The original source article explaining why organizations should balance AI-driven efficiency with employee experience, skills, risk management, and a people-first workforce strategy.

2. The Future of Jobs Report 2025

World Economic Forum

https://www.weforum.org/publications/the-future-of-jobs-report-2025/

Global employer research examining how AI, automation, economic change, demographics, and other forces may reshape jobs, skills, and workforce strategies through 2030.

3. The Impact of AI on the Workplace: Main Findings From OECD Surveys of Employers and Workers

Organisation for Economic Co-operation and Development

https://www.oecd.org/en/publications/the-impact-of-ai-on-the-workplace-main-findings-from-the-oecd-ai-surveys-of-employers-and-workers_ea0a0fe1-en.html

Research examining worker and employer experiences with AI, including productivity, job quality, working conditions, skills, and workplace concerns.

4. AI and Work

Organisation for Economic Co-operation and Development

https://www.oecd.org/en/topics/ai-and-work.html

An OECD research hub covering AI’s effects on employment, skills, productivity, worker well-being, and labor policy.

5. Artificial Intelligence Risk Management Framework

National Institute of Standards and Technology

https://www.nist.gov/itl/ai-risk-management-framework

The US government’s voluntary framework for managing AI risks to individuals, organizations, and society.

6. AI Risk Management Framework Resources

National Institute of Standards and Technology

https://www.nist.gov/itl/ai-risk-management-framework/ai-risk-management-framework-resources

Official implementation resources, profiles, guidance, and supporting materials for organizations managing AI risks.

7. Artificial Intelligence Risk Management Framework: Generative AI Profile

National Institute of Standards and Technology

https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence

A companion resource addressing risks specific to generative AI, including reliability, security, harmful content, privacy, and governance.

8. Artificial Intelligence and Employment

Organisation for Economic Co-operation and Development

https://www.oecd.org/en/publications/artificial-intelligence-and-employment_c2c1d276-en.html

Research examining the relationship between AI exposure, occupations, skills, and employment across countries.

9. Future of Jobs Report 2025: Skills Outlook

World Economic Forum

https://www.weforum.org/publications/the-future-of-jobs-report-2025/in-full/3-skills-outlook/

A detailed discussion of the skills expected to increase or decrease in importance as AI and other technologies reshape work.