Sort your week into green, amber and red
A 40-person landscaping company asks its office manager to use AI on a pile of 80 resumes. Before she opens a chat tool, she sorts her week into three colors.
15 min · You'll sort 12 HR tasks green, amber or red and know what stays humanYou'll sort 12 real HR tasks into green, amber and red, then sort your own week the same way.
Step 1 of 6: The idea
Step 1 of 6
The one idea this lesson is about, and why it matters.
None. This is the first lesson in the track, and it needs no paid plan and no HR software.
AI can draft the paper in HR work. A person decides anything about a person. That one rule sorts almost every HR task into three colors, and you can learn it in about ten minutes.
You don't have to like AI to try it once. It can be confidently wrong, so you read everything before it goes anywhere. It does not replace you either, because every decision about a person stays with a person.
The three colors. Green: AI drafts a document that you read and own, and no personal details about a named person go in. A job ad and a list of interview questions are green.
Amber: AI helps, but you check a law first, or you swap real names for placeholders (stand-in labels such as "Employee A"). A handbook policy and a work schedule are amber, because rules on hours and leave differ by place.
Red: a person decides, or the details must not go into a chat tool at all. Ranking applicants and choosing who gets disciplined are red.
Below, Marisol Duarte, the office manager of a 40-person landscaping company, sorts twelve tasks from her week.
The tool puts the survey summary in green. She moves it to amber, because one comment in a small crew can point to one person. Read her sort first, then do your own.
This track is training, not legal advice. Each lesson tells you what to check in your own place and when to ask a lawyer.
Why a person decides about a person. The EEOC (the US agency that enforces job discrimination law) says an employer stays responsible for a selection procedure. That holds even when a vendor made it.
A selection procedure is any test or screen used to decide who moves forward. The EEOC took down its two pages of AI guidance in 2025, but the laws those pages explained did not change. Courts are watching tool vendors too.
In Mobley v Workday, a federal judge let age-discrimination claims go forward against a company whose software screens applicants for employers. The applicants sue together as a collective.
As of Nov 19, 2026 the case is still in discovery (the stage where each side gathers evidence). No court has found the vendor liable, and the court order is linked at the end of this lesson.
A protected trait is a personal characteristic the law bars an employer from using against someone. Examples are age, race, sex, disability, religion, pregnancy, national origin and genetic information. Some states add more.
Four jobs you do not hand to AI. Each one gets its own lesson later in the track. Until then, treat all four as red.
1. Screening, ranking or scoring people. Screening means narrowing a pool of applicants in or out before a person has really looked at them.
New York City bars an automated screening tool unless it has had a bias audit (an independent check, covered in lesson 6). Candidates must also get notice.
2. Deciding who is hired, fired, paid, promoted or disciplined. A tidy AI recap of a manager's notes can drift into a recommendation, so you keep the decision and the reasons in your own words.
California signed a law on Sept 30, 2026, starting in 2027, that bars firing or disciplining someone based only on an automated system.
3. Guessing a protected trait. Do not ask a tool to guess someone's age, health, pregnancy, religion, race or sex from a resume, photo, voice or video.
The EU bans AI that infers emotions at work. The UK's ICO audit (linked at the end) found hiring tools guessing gender and ethnicity from names.
4. Pasting medical or leave details of a named person. That includes accommodation requests (asking for a change to the job because of a disability or religion) and complaints.
None of them go into any general chat tool, on a personal plan or a work plan.
US disability rules say medical information is kept in separate files and shared with only a short list of people. A chat tool is not on that list (our reading, so ask an employment lawyer). Anthropic's own support page, linked at the end, tells people not to put highly sensitive data into personal-account chats.
The vendors agree. Anthropic's usage policy (its rulebook for customers) calls employment a high-risk area. It asks for a qualified person who can change the AI's recommendation, and for telling the affected person that AI was used.
Google's policy bars automated decisions that harm people without human supervision in high-risk fields, employment included. OpenAI's developer guidance asks for human review of outputs and names high-stakes areas as especially critical.
Is your company small? Federal job discrimination law starts at 15 employees (20 for age claims). Your state may set its own limit, so check before you assume the rules skip you.
Who "counsel" is. HR counsel is an employment lawyer, meaning a lawyer who handles workplace law. If your company has none, you can hire one for a single consultation.
To find one, try your state bar's lawyer referral service (the bar is the group that licenses lawyers in your state). Or ask an HR hotline that comes with your insurance, payroll or PEO provider. A PEO is a company that handles payroll and HR for you, and a local employer association or chamber of commerce may also help.
Until you have one, leave AI out of any item this track sends to counsel. Hold documents such as a policy, a letter or a final review until a lawyer has read them.
A complaint, a safety problem, or a leave or accommodation request is different. You handle these by hand the same day, with the owner, because waiting can itself be a risk.
Handling it by hand. Follow these steps without AI and without a lawyer on the line.
To check your state, search "[your state] labor department" plus the topic, and look for a page address ending in .gov. This is training, not legal advice, and the pages above change.
Green you draft and own, amber you check first, and red a person decides.
That is the idea. Next you watch it work.
Step 2 of 6
See one real prompt do one real job, then see which part of it made that work.
You type this
I do HR for a 40-person landscaping company. Sort these 12 tasks into green, amber or red. Green: you draft a document I read and own, with no personal details about a named person. Amber: you can help, but I must check a law, or swap names for placeholders, first. Red: a person decides, or the information must not go into a chat tool. Give one line of reasoning for each. Do not do any of the tasks. 1. Draft a job description for a crew lead from my list of duties 2. Write a list of interview questions every candidate will get 3. Make a first-week checklist for a new hire 4. Write a friendly note announcing holiday hours 5. Draft a remote-work policy for the office staff 6. Turn my notes on one employee's year into clear review wording 7. Summarize the anonymous comments from our staff survey 8. Build next month's crew schedule from availability 9. Rank 80 resumes and pick the top 10 to interview 10. Decide which employee to discipline for missed deadlines 11. Read an employee's doctor's note and tell me what accommodation to give 12. Watch the video interviews and tell me which candidates seem healthy and energetic
Your AI tool does this
Your AI tool goes next.
Step 3 of 6
Write one part of the prompt yourself, then compare with one good answer.
Your rule for red
Two things make a task red. One is about who decides, and one is about the information.
Your answer is saved with your progress, so leave out anything private.
Try it before you go on.
Getting it wrong first is what makes it stick.
Step 4 of 6
Do the task on your own work, in your own tool.
List 12 tasks from your last two weeks of people work, with no names. Sort them with the rule above (in a chat tool or on paper). Argue with at least one color, and write down what moved it.
Go do it for real.
Come back with a result.
Leave out anything private. This uses one of your Course Council messages.
Step 5 of 6
Fill in your own card, then save it to your workbook.
Your takeaway
Loading your card…
Save your card before you go on. It goes in your workbook.
Step 6 of 6
Say it from memory first, then check it.
Pulling this out of your own head now is the part that actually moves it into long-term memory. Re-reading only feels like it does.
Without scrolling up, say the one rule of this lesson, and name the four jobs you never hand to AI. Then pick one task from your own week and say its color.
The rule: AI drafts and organizes, a person decides about a person. The four: screening or ranking people. Deciding to hire, fire, pay, promote or discipline. Guessing a protected trait. Pasting medical or leave details of a named person. If your task color came with a one-line reason, you have done it.
Sort a task. You can sort an HR task into green, amber or red and say why in one line.
AI drafts and organizes, and a person decides about a person. Sort every task green, amber or red, and treat screening, deciding, guessing protected traits and pasting medical details as red.
You do not need to remember the colors. Paste your list and ask: sort these green, amber or red using my rules, with one line of reasoning each, and do none of them. It is a quick first pass, and you decide whether you agree.
That is the lesson. Mark it complete, and its question comes back to you in three days.