You have a list of tasks from 3.2.3. To see where tools matter, you need two numbers for each task. The first is how far a tool reaches into the task today. The second is how far the four questions from 3.2.1 say a tool should reach. The difference is the gap.
Number one: reach
The reach scale
None.No tool does any of this task.
A small part.A tool helps with a small piece, such as a lookup or a form field. A person does the rest.
A big part.A tool does a big part, and a person finishes, fixes or decides.
Most of it.A tool does most of it. A person mostly samples the results or handles exceptions.
Reach is about what a tool does today, not what it might do someday. And every reach score above 0 needs evidence you can name: the worker told you, a published source says so, or it's your own guess. Guesses are allowed. Hiding them isn't.
Reach lab · invented tasks · nothing here is saved
“A scanner reads each paper form and fills in the fields. The clerk checks only the fields it marks red.”
“A designer asks an AI tool for ten rough logo ideas, picks one, and redraws it by hand.”
“A tool writes a first draft of every reply to customer emails. The agent edits each one before it goes out.”
“A shop owner counts the stock on the shelves by hand every Monday. No tool is used.”
“A chatbot answers every "Where is my order?" message. A manager reads a sample each week.”
“A mechanic's diagnostic tool narrows the problem to two likely parts. The mechanic tests each one before ordering.”
Correct so far: 0 / 6
Number two: how far it should reach
In 3.2.1 you answered four questions about a task and got a hand-off level from 1 to 4. The level says how much of the work a tool should carry. Here it becomes a ceiling.
From level to the most a tool should do
Level 1: AI does it.A tool should do up to reach 3 (most of it).
Level 2: AI drafts, a person reviews each.A tool should do up to reach 2 (a big part).
Level 3: AI assists, a person decides.A tool should do up to reach 1 (a small part).
Level 4: A person does it.A tool should do up to reach 1 (a small part).
ExampleOne task, start to finish
An invented clerk at an invented clinic enters insurance details from paper forms. Billing software reads each form and fills the fields, and the entries go straight to the insurer. No one reviews them.
Step 1 — Reach
The software does most of it. Evidence: the clerk said so. Reach 3.
⚠️A pressure point is not a prediction. It says a tool is doing more than caution would suggest. It does not say anyone will lose a task, and it does not say a tool should be removed. Room for help is the same: the questions would allow more, which isn't a recommendation to add a tool.
Put them together
Each case below shows a reach and the four answers. Work out the level, then the most it allows, then the gap, and name the category.
Gap lab · invented tasks, with four-question answers · nothing here is saved
Reset forgotten passwords after checking who is asking.A tool checks the person's answers and resets the password. A technician reviews the exceptions.
Reach 3 (Most of it) · Routine or judgment? Routine · How costly is a mistake? Costly, can be undone · Does it depend on trust? Doesn't depend on it · Who checks the AI's work? Only a sample or complaints
Decide which loan applicants get a second look.A tool ranks the applicants, and the officer almost always follows its ranking.
Reach 3 (Most of it) · Routine or judgment? Judgment · How costly is a mistake? Costly, hard to undo · Does it depend on trust? Matters some · Who checks the AI's work? No one checks
Order office supplies when stock runs low.The software suggests the order and a clerk clicks approve. A wrong order costs little.
Reach 3 (Most of it) · Routine or judgment? Routine · How costly is a mistake? Cheap to fix · Does it depend on trust? Doesn't depend on it · Who checks the AI's work? A person reads each result
Greet patients and check them in.A tablet takes signatures. A person greets everyone and fixes problems.
Reach 1 (A small part) · Routine or judgment? Mixed · How costly is a mistake? Cheap to fix · Does it depend on trust? Depends on it · Who checks the AI's work? A person reads each result
Write a first draft of next week's shift schedule.No tool is used. The manager builds it by hand every Sunday.
Reach 0 (None) · Routine or judgment? Mixed · How costly is a mistake? Cheap to fix · Does it depend on trust? Matters some · Who checks the AI's work? A person reads each result
Calm an upset customer at the counter.A tool suggests apology wording. The person leads the conversation.
Reach 1 (A small part) · Routine or judgment? Judgment · How costly is a mistake? Costly, can be undone · Does it depend on trust? Depends on it · Who checks the AI's work? No one checks
Correct so far: 0 / 6
Why there's no single “score” for the job
It's tempting to add everything up into one number. Don't. A made-up number looks precise and hides the choices behind it. Counts and lists are honest: one pressure point, three matched, three with room for help, and here are the tasks. Someone can check that.
💡This is a simplified teaching model. Real decisions have more than four questions, and real tools change. The model is useful for one thing: it makes your reasoning visible, so someone else can disagree with a specific score.
Principles of AI · 3.2.4Concept⚠ Watch Out
Published Data, Read Carefully
projection = an estimate for a whole occupation, not a forecast for your worker
One worker is one worker. To see how your findings fit a wider picture, you compare them with a published source. Two free ones are O*NET OnLine (onetonline.org), which lists tasks and other details for many occupations, and the Occupational Outlook Handbook (bls.gov/ooh) from the Bureau of Labor Statistics, which describes what workers do and gives a ten-year outlook.
Compare tasks
Below is an invented description, so no real page is copied here. Check each task a worker told us about against it.
An invented published description · “Dental Front Desk Scheduler”
Schedule and confirm patient appointments
Greet patients and check them in
Collect insurance information and verify coverage
Answer patient questions about billing
Maintain patient records
Compare lab · an invented worker's tasks · nothing here is saved
“Confirm next week's appointments by phone”
“Greet patients and check them in”
“Fill cancelled slots from the waiting list”
“Enter insurance details and flag missing forms”
“Order office supplies when stock runs low”
“Calm upset patients at the desk”
Correct so far: 0 / 6
What a match tells you, and what it doesn't
A task that isn't listed isn't wrong.A summary leaves things out, and your worker's workplace has its own tasks.
A listed task your worker never mentioned isn't an error either.Maybe someone else does it, or the interview didn't reach it.
Write the title exactly as the source lists it,and the date you looked. Pages are updated.
Read an outlook sentence
An outlook gives an expected change in the number of jobs in an occupation over about ten years. It's built from past data and assumptions, so it's a projection. It's a useful picture of a whole occupation across the country. It is not a forecast for one workplace, and it doesn't say which tasks change.
Three cautions
Different things are measured.A projection counts jobs. Your table is about tasks inside one job. They can fit together without one proving the other.
One worker is one worker.One interview can't speak for an occupation. Say so.
Copy it exactly.Copy the sentence word for word, with the years. Don't turn “change little” into “safe.”
When you decide how your findings fit the outlook, you have four honest choices: seem to fit, partly fit, seem to conflict, or can't compare directly. Then you give a reason. Try it on an invented outlook, “the number of scheduler jobs will change little,” and an invented worker with one pressure point.
Fit lab · claims about the invented outlook · nothing here is saved
“The invented outlook says the number of scheduler jobs will change little. It doesn't say which tasks a tool does.”
“The outlook proves tools aren't affecting this job.”
“A pressure point in one task is consistent with a stable number of jobs.”
“Because the number of jobs is stable, the worker's tasks won't change.”
“Every dental office uses the same billing software as the one the worker described.”
“The outlook and the worker's table disagree about this job.”
Correct so far: 0 / 6
🔑A pressure point in one task and a steady number of jobs can both be true. A task can change while the number of people doing the job stays about the same, and the other way around.
Principles of AI · 3.2.4SkillExample
Writing a Brief That Stays Inside the Evidence
brief = one page that says what you found, and what you don't know
A career brief is for a reader who hasn't seen your table. In one page, it should let that reader repeat what you found. Your project page assembles the brief from your own entries, so every sentence in it traces back to something you wrote.
What the brief holds, in order
The job and the source.What it is, and whether it came from an interview or a description.
One task where a tool reaches.With the evidence and the gap.
One task that stays with a person.And why.
The totals.Counts, and how many scores are guesses.
Published data.The source, the date, the match counts, the outlook sentence, and how your findings fit.
A skill to build and a first step.One you can start on.
What you don't know.One worker, a few tasks, a simplified model.
An invented example
Every person, office, tool and number below is invented, and so is its “published source.” The page builds it from the same data as the project does.
CAREER BRIEF: dental office scheduler (invented)
Source: one interview with Worker A.
The job: A scheduler keeps the appointment book full and the front desk running, and handles patients, phones and forms.
Where tools reach: "Enter insurance details and flag missing forms". A tool does most of it. Evidence (the worker said): The software fills in most fields from the form photo, and Worker A checks only the ones it marks red. The four questions point to level 2 ("AI drafts, a person reviews each"), so this task is a pressure point.
What stays with a person: "Calm upset patients at the desk". The four questions point to level 4 ("A person does it"), because each case needs judgment, it depends on trust, and no one would catch a mistake. A mistake costs the patient's trust, and nobody else would catch it, so a person stays in charge.
Across 7 tasks: 1 pressure point, 3 matched, 3 with room for help. 1 of 6 reach scores above zero is a guess of my own. Pressure points in time-heavy tasks: "Enter insurance details and flag missing forms".
Published data: I checked an invented guide for "Dental Front Desk Scheduler" (accessed October 2026). Of my 7 tasks, 1 listed there, 3 listed in different words, and 3 not listed. It lists 1 task I didn't have. The source says: “Employment is projected to change little over the decade (invented example).” (2024 to 2034). My findings partly fit with this. The invented guide expects little change in the number of jobs, and my table shows one pressure point in a time-heavy task, which a count of jobs wouldn't show.
If I were starting in this field: I would build checking a tool's output against the paper form (Steer). First step: Double-check the flagged list against the paper forms once a week.
“The tool is fast, but I'm the one who catches what it misses.” (Worker A)
What I don't know: I talked to one worker at one office about seven tasks, so this may not match other offices, and one of my reach scores is only a guess. This rests on one worker, 7 tasks, and scores from a simplified model. A projection describes a whole occupation, not one workplace, and it is not a forecast.
383 words · the limit is 420.
Check the claims
Some sentences report what the evidence shows. Others promise something nobody can know. Sort these lines from invented briefs.
Claim checker · lines from invented briefs · nothing here is saved
“In my notes, a tool does most of insurance entry, and the worker checks only the fields it flags.”
“This job will be gone in ten years.”
“Receptionists are safe from AI.”
“One worker told me this, so it may not match other workplaces.”
“The projection proves tools are taking the work.”
“Three of my seven reach scores are my own guesses.”
“Everyone in this field should learn to code.”
“I'd build the skill of checking a tool's output against the paper form, and start by double-checking the flagged list each week.”
Correct so far: 0 / 8
Words that usually go beyond the evidence
“Will be replaced,” “will disappear,” “safe from.”Predictions and promises. Say what you found today.
“Proves.”One worker and one table don't prove.
“Everyone should,” “all workers.”You looked at one job.
“Because of AI.”A fit isn't a cause.
Skill and first step
Pick a skill from your own table: perhaps something a person still does where a tool reaches far, or something the person-held task needs. Name its strand (Make, Know, Connect, or Steer) and a first step you could take this month that doesn't depend on anyone else. That is advice about a skill. It is not a promise about a job.
💡Access isn't equal. Not everyone could interview a worker, and not everyone has the same time or tools. The job-description path is a full path, and its limits go in the brief the same way. Nobody should lose credit for the people they happen to know.
Optional: a chatbot as a claim checker
If your teacher allows a district-approved chatbot, you can ask it to flag sentences in your finished draft that go beyond your evidence. Don't paste names, your school, or private details. Chatbots tend to be encouraging, so ask it not to praise. Then check its answer against your table yourself. It can't see your worker, and it can't check your source. Don't let it add numbers or facts.
Here is the draft of my one-page career brief. Don't rewrite it and don't praise it. For each sentence, tell me whether it reports what my table or my source shows, or whether it goes beyond that (a prediction, a promise, a claim about everyone, or a cause I haven't shown). Quote the sentence, say which, and say what I could write instead using only my own evidence. Then list anything the brief says that my table does not show.
[paste your brief here. No names, no school, no private details]
Your worksheet
🩻 Task X-Ray Worksheet⬇ PDFThe Google Doc copy is on the project's start page.