How AI Changes Tasks And Jobs

Imagine a bicycle repair shop trying an AI assistant for customer messages. Maya coordinates the service desk. The assistant quickly drafts a friendly reply about when a bicycle will be ready. The manager asks whether this means the shop needs fewer coordinator hours. To answer, we will follow that customer request through the shop and see how the work fits together. This is Lesson two, How AI Changes Tasks And Jobs. By the end,

you should be able to explain how a change in one activity reaches people’s roles, and what you would need to investigate before predicting an employment outcome. The customer asks, can I collect my bicycle on Friday? The assistant drafts a confident yes. Maya opens the repair record. Inspection is still pending, and the mechanic has yet to confirm which part is needed. She replaces the promise with a message explaining when the customer will receive an update.

Producing the reply is one task: an activity with a purpose and an output. Making sure the shop gives accurate information is a responsibility. That obligation continues through the draft, the check and the message the customer receives. Maya now needs information that the assistant’s draft could not supply. She asks the mechanic about the inspection, contacts the supplier about availability and checks the customer’s preferred contact method. These relationships make the service possible.

The mechanic can confirm the technical work; the supplier can confirm delivery; the customer can explain a collection constraint. Maya connects their answers. A job brings together tasks, responsibilities and relationships. Following one request reveals how those parts depend on one another. There is also a decision to make. Suppose the supplier offers a quicker delivery at an extra charge. The shop has to decide whether to pay it and what commitment it can make to the customer.

An assistant could organize the options. The shop must specify who can authorize the spending and promise a collection time, and who will explain the decision if it goes wrong. That is accountability. In this imagined request, the manager holds off on extra delivery spending until the part is identified. Maya explains that inspection and availability still need confirmation and gives the customer a time for the next update.

A useful task map records this authority alongside the activities. It lets us see which part of the role a tool can perform and where a decision still needs an authorized answer. With the customer’s update sent, the manager looks back at the work. The assistant helped prepare wording. Maya checked the status, coordinated answers, handled an exception and followed the authorization rule. Some of those activities might also change as the system develops.

For example, a reliable connection to current repair records could reduce manual lookup. That would need testing against real requests, including missing information and conflicting updates. The task map is a record to revise as the process changes. Trying that connection creates work of its own. Someone has to keep repair statuses current, decide which information can enter a customer reply and test what happens when a delivery date is missing.

Maya and the mechanic may need a clear way to flag an unreliable draft. A manager needs to assign who handles escalations. These activities use staff time and skills. Record who takes them on. If a shorter drafting step shifts checking to the mechanic, the change has reached another person’s workload. The manager can now ask a better question: what happens to the complete service once the tool and the surrounding work are in place?

Suppose a trial establishes that the shop can handle a customer request with less total staff effort while meeting its service standard. The shop could use the released capacity to give customers more updates and help with unusual requests. Maya’s role could then contain more coordination and personal support. With service volume held fixed, the manager might instead reduce scheduled drafting hours. That choice could reduce paid work unless other duties or demand take its place. The verified capacity gain opens these alternatives;

the shop’s service goals, finances and decisions determine which it pursues. Before choosing, the shop also has to consider adoption. It needs money for the application, integration and training. Privacy obligations shape the records it can use. Customers may value a conversation when a delay disrupts their plans. Staff may identify failure cases or propose a better division of work. A demonstration of a good reply begins this investigation. Employment consequences emerge through adoption, the redesigned workflow,

demand for the service and decisions about how to use any gain. This chain helps us read evidence about AI and jobs. In May 2025, the International Labour Organization published a refined global index of occupational exposure to generative AI. It combines task information, expert input and model predictions to classify potential exposure. For the shop, that kind of method asks where capabilities could interact with the content of a role.

Its exposure categories describe technical potential under the study’s definitions. Estimating how many people will remain employed requires the later links we have just followed: adoption, work arrangements, service demand and organizational choice. Employment records help investigate those later links. The August 2026 revision of Stanford researchers’ Canaries in the Coal Mine paper examines payroll data through June 2026. It reports weaker employment patterns for workers aged twenty-two to twenty-five

in more exposed occupations, with reduced hiring an important part of the pattern. The researchers describe the findings as early descriptive indicators. Education, earlier trends and the payroll sample’s coverage affect the comparison. For our shop, this suggests questions about entry opportunities as well as existing jobs: will a new coordinator still be hired, and how will that person learn the work? A payroll pattern can guide that investigation, while a claim about what caused a particular staffing change needs further evidence. Try the Learning Check.

Someone says, a quarter of jobs exposed means a quarter disappear. What is wrong with that conclusion? Explain what exposure measures and name at least one additional mechanism you would need to investigate before making an employment claim. Pause and give your answer. Exposure measures potential interaction with task content under a study’s method. Turning it into a disappearance rate skips adoption, the work that remains or is created, demand and organizational decisions.

You could investigate whether an application is actually used, how the complete role changes and whether the organization expands service or reduces hours. An employment claim also needs actual outcome evidence with its population, period and method attached. Return to Maya and the manager. We can now replace the original question about a fast reply with a concrete investigation. Map the customer request from first message to a usable answer.

Identify the tool’s activity, the coordination and checking around it, any new work and the people who can authorize commitments. Then compare complete service effort and quality in actual use. Ask how the shop would use any released capacity and what that would mean for staff hours, hiring and customers. This gives the manager a way to examine a proposed role change and gives staff a way to challenge assumptions with specific work and evidence. For Practice, map a different fictional role: a community library activity organizer.

Use your worksheet to identify at least four tasks, possible AI assistance, remaining human duties and new work. Include one relationship and one decision for which accountability matters. Choose an adoption condition to investigate, and state the evidence you would need before making an employment claim. Keep the response fields for your attempt; the separate key offers an example for comparison. The shop’s assistant changed an activity inside a connected service.

Its consequences depended on what happened around that activity and how people used the resulting capacity. You can now trace that pathway from task to role to an employment question. One issue remains: how does making an activity cheaper change demand for related work? The next lesson follows that question through substitution, complementarity and demand.