New research from the Notre Dame–IBM Technology Ethics Lab finds that AI adoption creates additional verification, training, and coordination responsibilities, with middle managers and some women reporting a disproportionate share of the burden.

WHAT’S HAPPENING

Artificial intelligence is being introduced into workplaces with the promise of saving time, improving productivity, and reducing repetitive work.

But new research suggests that some companies may be overlooking an important part of the equation: the additional work required to make AI function effectively.

Research involving the University of Notre Dame and IBM’s Technology Ethics Lab examined how employees and managers experience AI adoption inside organizations.

The findings highlight a growing category of responsibilities described as “glue work.”

Glue work consists of activities that keep organizations operating smoothly but are often overlooked when companies measure productivity.

As businesses introduce AI tools, employees may find themselves performing new tasks, including:

• Verifying AI-generated information: Checking reports, summaries, recommendations, and other outputs for accuracy.

• Correcting mistakes: Identifying and repairing errors produced by AI systems.

• Training coworkers: Helping employees learn how to use unfamiliar technology.

• Coordinating implementation: Making sure AI tools work with existing business processes.

• Maintaining accountability: Determining who is responsible when AI produces incorrect or incomplete work.

• Translating management objectives: Turning executive decisions about AI adoption into practical procedures employees can follow.

The researchers conducted workshops in July 2026 involving middle-level professionals responsible for implementing AI and executives overseeing AI adoption.

Participants described additional responsibilities that were not always formally recognized in job descriptions, staffing decisions, or productivity calculations.

Middle managers were particularly important because they frequently found themselves responsible for making AI systems work while continuing to perform their existing management duties.

Some participants also reported that women were taking on a disproportionate share of coordination, training, and support responsibilities.

The research does not establish that women across the workforce universally perform more AI-related work. Its findings identify a concern raised by participants that warrants broader investigation.

Another tension emerged.

Some organizational leaders viewed AI as a technology that could eventually reduce the need for management positions.

Yet those same organizations depended on managers to introduce the technology, supervise its use, and correct problems.

The central finding is that AI may eliminate certain tasks while creating additional responsibilities that businesses do not always measure or reward.

WHY IT MATTERS

The financial argument for adopting AI generally depends on efficiency.

A business purchases AI software, employees complete tasks faster, and management expects productivity to improve.

That calculation appears straightforward.

But it can become misleading when organizations measure only how quickly an AI system generates an initial result.

Consider a company that previously required two hours to prepare a detailed business report.

An AI assistant might produce a draft in 20 minutes.

Management could conclude that the company has reduced the work by 100 minutes.

But suppose a manager must then spend another 40 minutes reviewing the report, verifying numbers, correcting inaccuracies, and confirming that it meets company requirements.

The actual process now takes one hour rather than two.

The business still benefits, but the productivity improvement is smaller than the initial calculation suggested.

There is another complication.

The employee using AI may receive credit for completing the assignment quickly, while the manager performing the verification absorbs additional work.

The business saves time in one department or position while shifting responsibilities to another.

If the company does not track those transferred responsibilities, it may misunderstand where productivity gains are occurring.

This matters because AI-generated material often requires human oversight.

Financial figures must be accurate.

Customer communications must reflect company policies.

Legal documents require appropriate professional review.

Technical recommendations must be evaluated before implementation.

Employees also need training, and existing procedures may require adjustments.

These responsibilities involve labor, time, and sometimes additional technology expenses.

The research challenges a common assumption that AI productivity can be measured simply by comparing the time required to complete an individual task before and after automation.

Instead, companies need to examine the total cost of producing an acceptable final result.

That includes AI generation, human verification, corrections, employee training, technical support, and ongoing maintenance.

The distinction becomes especially important when companies expand AI across multiple departments.

A tool that works efficiently for one employee may generate substantial coordination requirements when introduced across an entire organization.

Managers may need to resolve disagreements about acceptable AI use, determine which information employees can enter into systems, and establish procedures for handling mistakes.

Organizations may also need to develop rules governing privacy, security, accountability, and quality control.

Those activities can be essential even when they do not directly generate revenue.

The study also raises an important workforce question.

Who performs the work required to make AI adoption successful, and are those employees receiving appropriate recognition?

If coordination responsibilities fall unevenly on certain employees, organizations may unintentionally create additional workload without adjusting compensation, staffing, or performance expectations.

The concern involving women is particularly relevant because administrative support, workplace coordination, and informal assistance have historically not always received the same recognition as directly measurable production activities.

The workshop findings suggest that AI adoption could reproduce some of those patterns.

However, larger studies are needed to establish their prevalence and determine the causes.

For business owners, the practical implication is clear.

A successful AI implementation should not be judged solely by whether the software performs its assigned function.

It should be judged by whether the organization operates more effectively after accounting for all the work required to use it.

WHO BENEFITS

– Businesses measuring total productivity: Can make better investment decisions by accounting for AI-generated work, human verification, training, and ongoing support.

– Middle managers: May gain stronger justification for appropriate staffing, training resources, and recognition of additional implementation responsibilities.

– Employees performing previously unrecognized tasks: Could benefit if organizations begin formally measuring and compensating coordination, verification, and support work.

– AI implementation consultants: May gain opportunities to help businesses redesign workflows and calculate the actual costs and benefits of automation.

– Employee training providers: Could see increased demand for practical instruction on responsible AI use, output verification, and workplace integration.

– AI software developers: May benefit from demand for systems that require less correction, provide clearer audit trails, and integrate more smoothly into business operations.

– Small and midsize businesses: Can avoid expensive implementation mistakes by evaluating staffing requirements before introducing AI across their organizations.

– Organizations involving employees in technology decisions: May develop more effective AI implementations by identifying operational problems before they become widespread.

WHO LOSES

– Businesses relying on exaggerated productivity projections: Risk investing in technology that provides smaller financial benefits than originally anticipated.

– Organizations ignoring hidden labor costs: May experience employee burnout, operational confusion, or disappointing returns on AI investments.

– Employees assigned additional responsibilities without support: Could face heavier workloads without appropriate training, staffing, or compensation.

– Managers responsible for poorly planned implementations: May become accountable for AI performance without receiving the resources needed to manage it.

– AI vendors promoting unrealistic automation claims: Could face increased scrutiny as customers demand evidence of total operating savings rather than isolated task improvements.

– Businesses eliminating positions before understanding new workloads: Risk creating operational gaps when necessary oversight and coordination responsibilities remain.

– Organizations with weak performance measurement systems: May struggle to identify whether AI is actually improving productivity or simply redistributing work.

– Employees who receive inadequate AI training: Could face increased pressure to use unfamiliar systems while remaining responsible for errors they do not fully understand.

WHAT HAPPENS NEXT

The next important development will be determining how widespread these hidden AI-related responsibilities have become.

The Notre Dame–IBM research identifies a meaningful organizational problem, but its workshop-based findings do not establish the percentage of businesses experiencing it.

Larger workplace studies will be needed to measure the additional labor created by AI implementation.

Researchers will also need to examine whether certain employees consistently absorb more of that work and how responsibilities vary by industry, occupation, company size, and management structure.

For employers, the findings suggest several practical changes.

Measure the complete workflow

Companies should evaluate how long it takes to produce an accurate, usable result—not merely how quickly AI generates an initial response.

That means including review time, corrections, training, support, and coordination.

Identify who performs the additional work

Organizations should determine whether AI-related responsibilities are being assigned formally or absorbed informally by employees.

If a manager spends substantial time supervising AI-generated work, that effort should be included in operating calculations.

Review staffing assumptions

Businesses should avoid assuming that faster task completion automatically justifies reducing staff.

Some responsibilities may disappear, while others emerge.

The total workload must be measured before making staffing decisions.

Improve training

Employees need clear guidance about what AI systems can do, where they make mistakes, and when human review is necessary.

Better training could reduce avoidable corrections and confusion.

Establish accountability

Companies should define responsibility for AI-generated information, customer communications, and operational decisions.

Clear ownership can prevent problems from being transferred between departments.

Evaluate the quality of AI systems

An AI tool that generates results quickly but requires extensive correction may offer less value than a slower system that consistently produces accurate work.

Reliability should be included in technology purchasing decisions.

Monitor employee workload

Organizations should examine whether AI implementation increases pressure on certain employees or departments.

This includes investigating whether coordination and support responsibilities are being distributed fairly.

Calculate the actual return on investment

The relevant financial comparison is not simply the cost of an AI subscription versus the estimated hours it saves.

Businesses should consider implementation expenses, software costs, ongoing maintenance, employee time, and measurable improvements in operations.

The research also creates an opportunity for companies providing AI services.

Customers may increasingly demand implementation support rather than software alone.

That could include workflow assessments, employee training, quality-control procedures, and ongoing performance measurement.

For AI providers, the ability to demonstrate reliable business outcomes may become a competitive advantage.

A system that requires fewer corrections and less management oversight could deliver greater value even if its subscription price is higher.

The same principle applies to organizations considering whether to build custom AI applications or purchase existing products.

They should evaluate the entire operating process, including what happens when the AI makes mistakes.

Over time, businesses may begin paying more attention to the distinction between task-level efficiency and organization-wide productivity.

Task-level efficiency measures how quickly an individual activity is completed.

Organization-wide productivity examines whether the business produces better results with the resources available.

Those outcomes are related, but they are not identical.

A company can automate hundreds of activities without improving overall performance if the technology introduces substantial new coordination costs.

Conversely, an AI system that saves only a modest amount of time on individual tasks may provide significant organizational benefits if it reduces errors, improves consistency, or simplifies communication.

The strongest implementations will be those that measure both sides of the equation.

The larger shift is clear: Artificial intelligence is changing not only how work gets completed but also where responsibility for that work resides. Research from Notre Dame and IBM’s Technology Ethics Lab suggests that managers and other employees may be absorbing additional verification, training, and coordination duties that traditional productivity measurements overlook. As businesses invest more heavily in AI, the companies that achieve meaningful returns will be those that measure the complete cost of automation, recognize the people making it work, and focus on improving overall operations rather than simply generating results faster.