Try each exercise first, then open its model answer.
AI in your job (role tracks) · Workbook
Reviewed and published, version 1.0.0. Big Widget Company and its people are made up. The model answers are examples of what a good answer contains, not the only right answers. A person reviews your work at 100.P. An AI does not.
What not to write here. Do not paste real prompts, AI answers, customer details, supplier data or other personal or secret information into these exercises. Use made-up or general examples. If your workplace lets you keep notes on a real prompt, answer and check, keep them in a separate work record that your workplace controls.
Exercise 1 (100.R-a): one good use, one bad use, one question
Take the track for your own team, or the home track if you use AI only at home.
- Pick one row from your track, or write a new row for your team or household.
- In your own words, write:
- Good use: AI doing a job it is good at, with safe information, checked by a named person who owns the result.
- Bad use: the same tool on a job it is bad at, or with secrets it should not see, or with nobody checking, or used to fool people.
- Get-to-yes question: the one question, asked before you start, that targets the check this job is most likely to fail.
- Say why the question belongs to your team and not only to IT. Name one thing your team must contribute or confirm, even if IT also knows it: which facts are secret, what a wrong answer would cost, or who should check.
- Your question is necessary, not sufficient. Go through all four 100.9 checks: the task fits the tool, the data is allowed, a named person owns the result, a written rule allows the action. For each, write yes, no or "don't know yet". If any is no or "don't know yet", write the gate answer (wait for a person) and who will decide.
- Write the use as a rule with the 100.9 template: This AI may [do what] for [whom] only when [condition]. Otherwise it [falls back to what: only this far, wait for a person, or no]. [Named person] owns it. We keep [this receipt]. The receipt records the decision, not the content.
Model answer
An example for the customer service track at Big Widget Company (made up):
- Good use: an AI helper suggests replies; Sam, the agent, decides what the customer is told.
- Bad use: a chatbot answers customers alone, invents a policy, and nobody can reach a person.
- Get-to-yes question: Can it only give answers that match our written policy, and can every customer reach a person?
- Why it is our question: customer service knows which policies change often, what a wrong promise would cost the customer and the company, and which senior agent should check new reply templates.
- The other checks: task fits (drafting replies: yes); data allowed (the helper sees the open ticket only, and that is approved: yes); named owner (Priya, the team lead: yes); written rule allows it ("don't know yet", because the team has not written the rule down). Gate answer: wait for a person until Priya writes the rule and it is approved. A hold is not a quiet yes: if nobody decides, the answer is no.
- Rule: one workflow with two decision points, because drafting and sending are different actions.
- Draft: This AI may draft replies to customer emails for our service agents only when it uses the open ticket and our approved policy text, and the draft stays private to the agent. Otherwise it does not draft (no).
- Send: A reply may be sent to a customer only after the agent handling the ticket has read and edited it and chosen to send it. The AI never sends on its own. Otherwise the reply waits for a person.
- Priya, the customer service team lead, owns both decision points. We keep a receipt of each approval: what was asked to run, the decision, the owner and the date, with no customer details.
A good answer:
- names a good use and a bad use that are recognisably the same kind of tool, separated by a choice people make before the AI acts;
- asks one focused question that targets the check this job is most likely to fail, and does not treat a yes to it as permission on its own;
- names a person, not "a human in the loop" or "the AI";
- sends anything the AI cannot undo, such as a payment, a public post or a change to live data, to a person before it happens;
- keeps prompts, AI answers and personal data out of the receipt.
Exercise 2 (100.R-b): where someone else's AI mistake lands
Part A: one documented route (everyone does this). Pick one of the four routes in the lesson's table: through a company you deal with, through a decision about you, through what you read, or through a product you use. Take the role it names: customer, applicant, shopper or owner. You are explaining how the mistake could land on someone in that role. Do not claim it happened to you. Write:
- whose AI or automation made the mistake;
- what output carried it;
- who it reached, and what it cost or could cost;
- who should be told.
Part B: one other team. Read one other team's track. Name one way an AI mistake there could land on you, and who you would tell. If you look and find no route, say so and say why. Do not invent one. "I don't know yet" is also an honest answer, as long as you name who you will ask.
Model answer
Part A, an example using the route "through a decision about you" (role: applicant):
- Whose automation: iTutorGroup's tutor application software. According to the EEOC's lawsuit, it was programmed to automatically reject female applicants aged 55 or older and male applicants aged 60 or older.[s2] The EEOC's release does not mention AI, so this is not evidence that generative AI caused the rejections.
- What carried it: the automatic rejection of the application.
- Who it reached: applicants, who did not build or switch on the software. The EEOC said more than 200 qualified applicants in the United States were rejected because of their age, and the company agreed to pay $365,000 to settle.[s2]
- Who should be told: the person at the company who owns its hiring software, who can stop the screening and have every rejection reviewed. A good answer names a role like this. Which outside body to go to is a legal question this module does not cover.
Part B, two examples (Big Widget Company, made up):
- A route. I work in finance. IT gives an AI agent too much access, and it changes data finance uses to close the books. Finance finds the error at month end. I would tell the IT owner of that agent and my finance manager. The question I would ask IT: What can it change, and can we undo it?
- No route, said honestly. I work in the home track. I read the IT track's example of agents running in a safe test area with backups. As written, I found no way that use reaches me: its changes stay in the test area and touch nothing I use. That rests on the test area really being sealed off, so if I am unsure, I would ask the IT owner what the agent can reach.
A good answer:
- in Part A, names all four parts (whose AI, what output, who it reached, who should be told) and stays within what the documented case shows;
- does not claim the mistake happened to the learner;
- names a person or role to tell, not just "someone";
- in Part B, either traces a route carried by that specific AI use, or says honestly that no route was found and why, without inventing one. Finding no route to you does not mean the use is safe for everyone.