When AI Helps Workers Or Replaces Tasks

Imagine a community museum that helps visitors plan a trip. Staff answer questions about exhibitions, opening times and access. An AI assistant can prepare parts of a reply more quickly. The museum hopes to serve people who previously found planning a visit difficult. Its staff want to know what will happen to their work. This is Lesson three, When AI Helps Workers Or Replaces Tasks. We will follow a visitor request, then calculate how serving more visitors could change the amount of work the museum needs.

Rosa is planning a visit with her father. She asks for a short route with places to sit. The assistant turns the museum’s approved information into an initial reply. Leela, the visitor adviser, checks a temporary gallery closure and asks which parts of the collection interest them. She adjusts the route around their answers. Preparing initial wording takes less of her time. Her knowledge of the museum

and her conversation with Rosa help turn that wording into a useful plan. This gives us two relationships to examine. Substitution occurs when a capability replaces some use of another input: here, some of Leela’s initial wording activity. Complementarity occurs when a capability makes another input more useful or productive. Having a draft to work from could help Leela use her visitor knowledge to develop a suitable route. Both relationships can operate within the same request.

To identify them, follow the particular activity and ask how the other contributions change. The museum then tests routine opening-time questions. If the application can reliably use current information and deliver an acceptable answer, automation could cover more of those requests. Route planning may involve different information, preferences and exceptions. The boundary has to be tested for each service. It can change as capabilities, information systems and user acceptance change.

The work map should follow that evidence rather than freeze today’s division of activities in place. A documented support system shows one way assistance has been arranged. In the published 2025 study Generative AI at Work, an assistant suggested responses during customer-support conversations. Agents could edit or ignore them and remained responsible for the conversation. The study examined one firm’s workflow, including issues resolved per hour. For our museum, the useful comparison is the complete interaction:

which part can the application perform, and how does that change the adviser’s contribution to an accepted result? The following lesson will examine workplace study findings and the comparisons behind them in more depth. Suppose the museum’s trial shows that the complete planning service needs fewer staff minutes per accepted response at the same service standard. The director now has another decision. Should the museum keep the same number of requests,

offer the service more widely or use the time for difficult visits? A faster process changes what is feasible. Demand determines how much of the service people want to use, while the museum’s decisions and capacity determine how much it provides. Work through an imagined day. The museum completes twelve accepted responses, each requiring thirty labor minutes across the full service. Twelve times thirty gives three hundred and sixty labor minutes.

The redesigned service takes twenty minutes for each accepted response. At the same twelve-response volume, total labor becomes twelve times twenty: two hundred and forty minutes. The museum has released one hundred and twenty labor minutes under these assumptions. It can now decide what those minutes should support. The director proposes advertising the planning service through community groups. Suppose that brings the daily volume to twenty-four accepted responses. At twenty minutes each, the total becomes four hundred and eighty labor minutes.

Each response still requires less effort, yet the larger service uses more total labor than the original three hundred and sixty minutes. The lower effort per response and the higher total workload are both consistent with these figures. The changed assumption is the number of requests the museum completes. Why might more people use the service? An easier way to ask, a lower price or a more useful response can make a service attractive to people who previously went without it.

At this museum, community outreach and useful planning could help people consider a visit. The director would need actual request and visitor records to find out whether that happens. Other conditions could limit expansion: people may have little additional need, transport may be difficult, or the museum may lack staff capacity. Better production and wider use have to be connected through the actual setting. Those conditions change the options for workers and visitors. With volume fixed,

the director could assign released time to personal assistance or reduce scheduled work. With more requests, staff could handle a larger service and a different mix of difficult cases. Training may be needed for that mix. Visitors could gain access to support, and the museum must decide how to fund and maintain it. Staff and community groups can help shape that decision. The twelve-response and twenty-four-response calculations

make the alternatives inspectable; actual staffing and service records would show which path occurs. The same multiplication also tells us where to look beyond labor. For any measured resource, total use depends on the amount used for each response and the number of responses. The application’s computing charges and energy require their own records and units. More visitor requests could change those totals too. To explain employment or pay, we would additionally need hiring,

scheduled hours, roles, budgets and organizational decisions. A labor-minute calculation gives us one part of that investigation. Try the Learning Check. Why should we analyze both task effort and demand? Explain how less effort for each unit can coexist with more total work, and identify what must be checked before describing an employment or resource-use outcome. Pause and reason through your answer. Multiply effort per accepted unit by the actual number of units.

If the volume grows enough, total effort can rise while effort per unit falls. Check the full task boundary, the acceptance standard, the observed volume and the relevant resource measures. Then investigate how the organization responds through service provision, staff hours, hiring and other decisions. Those records connect the calculation to an employment or resource-use explanation. Return to Rosa’s request and the museum’s decision. The assistant can replace some wording activity

while making the adviser’s knowledge more useful. Testing the complete service reveals where that boundary works. The director then needs to examine use: twelve requests at twenty minutes releases time; twenty-four requests at twenty minutes creates a larger total workload. You can now explain why a faster task can accompany either a smaller workload or an expanded service, and identify the evidence and choices that distinguish those paths. For Practice, use the different fictional service in your worksheet.

Compare its original workload with a faster process under two supplied volume assumptions. Calculate total labor minutes for each. Identify an activity that might be substituted, a human input that might be complemented and evidence that would test the demand assumption. Leave the response fields for your own attempt and consult the separate key afterward. We have followed a change inside a request through to the scale of a service and its consequences for people. The next question is how much confidence to place in evidence that a new process works.

Lesson four examines workplace studies, their findings and the comparisons needed to decide whether a result is useful in another setting.