Question 1

What is the Work Atlas?

Accepted Answer

A map of the work people actually do, and how to point AI at it. It holds 1,949 careers, 500 human activities, the soft skills behind them, and the tools they run on — each with a ready prompt. It is not a list of prompts to scroll through: you find your own work first, and the prompt is what's waiting for you there.

Question 2

Why do I have to pick a starting point instead of just searching?

Accepted Answer

You can search — the box works fine. But most people don't know the exact words yet, so there are four doors into the same library, and you can pick whichever one matches what you already know: Your job — you know your role; start there and see its real tasks and tools. Your tool — you live in Acrobat or Photoshop; start from the software. A human skill — the part that is about people, not software. Something you do — an activity like negotiating or baking, whatever your job title.

Question 3

Do I need an account to use it?

Accepted Answer

No. Everything on the page is open — search it, browse it, copy any prompt. Nothing is gated, and nothing asks who you are before it helps you.

Question 4

How is this different from a prompt library?

Accepted Answer

A prompt library is a list of clever text. This is a map of work: Every prompt is attached to a real, evidenced task from a real occupation. Prompts are typed by what they are for — a drill, a diagnosis, a conversation, a reference. It tells you where AI cannot help, which no prompt list will do. It starts from your job, your skill or your activity — not from a keyword.

Question 5

Who is this for?

Accepted Answer

Anyone who works. It covers trades and crafts, healthcare, teaching, accounting, science, transport, care work and office work — not just desk jobs. If a job exists, it is meant to be in here.

Question 6

How do I find my job?

Accepted Answer

Three ways, in order of speed: Type it in the search box — autocomplete finds jobs, skills, activities and tools at once, and the shape on each suggestion tells you which kind it is. Use the wizard — answer two or three light questions and it narrows 1,949 careers to a handful. Browse the wall — it deliberately spreads across every family of work, so you can recognise yourself instead of scrolling an alphabet.

Question 7

What are the highlighted words in square brackets?

Accepted Answer

They are slots — the parts only you can fill: a name, a date, a number. Everything else is written for you. Replace the slots, paste the prompt, and the AI has what it needs to do the work rather than ask you twenty questions first.

Question 8

What do the different prompt types mean — drill, diagnostic, applied, and the rest?

Accepted Answer

A prompt is a tool with a purpose, and using the wrong one is why people conclude AI is shallow. There are six: Drill — you practice on a fake version of the real person (an angry customer, a strict examiner). It will not let you win easily. Diagnostic — you describe what went wrong, and it ranks the most likely causes, best guess first. Applied — you give your own details and limits, and it turns them into a finished answer built for you. Grounded study — you learn the real rules of a topic, including where the experts disagree. Conversation — a longer back-and-forth, not a single answer. Reference — a plain fact, asked directly.

Question 9

What do Execute, Improve, Decide and Become mean?

Accepted Answer

Execute — Do the immediate task, with fewer mistakes. Improve — Make the work easier for the next person to accept. Decide — Diagnose the situation when the right move is unclear. Become — Learn the pattern so the problem stops returning. Most people only ever use Execute. The value compounds higher up.

Question 10

Which AI should I paste these into?

Accepted Answer

Any capable assistant. The prompts are written for the work, not for a product — they name what you want done, never the button that does it. That is deliberate: the tool changes, the work does not.

Question 11

The prompt is close but not quite my situation. What do I do?

Accepted Answer

Edit it — it is a starting point, not a spell. What it carries is the shape of a good request: the context, the constraint, the output you want, and permission for the AI to push back. Change the specifics and the shape still works.

Question 12

What do XS, S, M and L mean on an activity?

Accepted Answer

XS — Exhausted in a handful of questions. Ask twice and you are asking the same thing. S — A finite rule-set. One good session gets you everything; then you just do it. M — Real theory and real diagnosis — but reality holds the grading pen. L — Unbounded, and the loop closes: the AI can be the other person, or the artifact.

Question 13

Why is negotiating rated so much higher than swimming?

Accepted Answer

Not because it is deeper — swimming has an enormous literature. It is because the model can be the other side of the table. With swimming, once the theory runs out you have to get wet, and nobody is watching your stroke. The test is never how important something is; it is whether the loop can close.

Question 14

Can AI actually see or judge what I did, or can it only talk about it?

Accepted Answer

Depends on the task. We call this "closing the loop" on the page — whether the AI can be part of the actual result, or only see it after the fact. Three levels: It can be the artifact — if the output is text (an email, a contract, code, an argument), it does the work with you. It can be the counterparty — the landlord, the defensive engineer, the frightened family. Infinite reps, no social cost. It can half-see — a photo of your bread closes part of the loop. Sound, feel and timing close none of it.

Question 15

You say AI cannot help with 71 activities. Why keep them?

Accepted Answer

Because saying so is the useful part. If the honest answer to ‘how do I get better at blinking’ is ‘you do not, it is a reflex’, then telling you saves you an afternoon. A map that only shows where the treasure is, and never where it is not, is not a map.

Question 16

What is the most common mistake with the M tier?

Accepted Answer

Mistaking a well-explained answer for progress. In M — baking, welding, lifting, swimming — the AI writes a superb twelve-week programme and has no idea whether you did it well. Bring it real data: photos, weights, times, temperatures. Every gram of real evidence turns something it cannot judge into something it can.

Question 17

Will the XS/S/M/L ratings on this page update as AI gets better?

Accepted Answer

Yes, and that is the point. Every capability jump promotes things. Voice already argues singing and speaking upward. A model that can watch you deadlift promotes half of the M tier. The rating has a shelf life by design — re-run it, and the difference is a map of what changed.

Question 18

Where do the tasks come from?

Accepted Answer

From occupational evidence, not imagination. Every task is confirmed by at least two independent sources — O*NET (the U.S. government's official database of what every job actually does), ESCO (Europe's equivalent), real job postings, and reference material — before a prompt is written for it. If only one source claims a job does something, we leave it out.

Question 19

How do you know which software a job uses?

Accepted Answer

From the occupational record of that job, not guesswork. That is how a Forensic Nurse Examiner ends up with Epic, a Licensed Practical Nurse with MEDITECH, and a CNC Programmer with GibbsCAM. We did not choose those; the evidence did. 668 distinct tools surfaced this way.

Question 20

What are the soft-skill prompts based on?

Accepted Answer

For each skill, we collect the real books and named methods that working professionals actually use — and, importantly, the places where those experts disagree with each other. A prompt built on a real disagreement makes you decide something. A prompt built on a plain definition just makes you read.

Question 21

What are the shapes next to everything?

Accepted Answer

Expressive shapes — dual-tone symbols chosen to mean the thing, never to decorate it. Every career has its own, matched semantically and unique on the page. They are how you recognise a job, a skill, an activity or a tool at a glance, before you read a word.

Question 22

Can I use this for medical, legal or financial decisions?

Accepted Answer

No. Some of the deepest activities here — healing, taxing, investing, prosecuting — score high because they are text-shaped and endlessly discussable, not because an AI substitutes for a professional. Depth is not authority. Use it to prepare for the conversation with the professional, not instead of it.

Question 23

Are all the prompts you generated published?

Accepted Answer

No. Around 9% are withheld because they name a software product inside the prompt, which breaks our own rule that a prompt names the work and never the button. They exist; they are simply not shipped. We would rather show fewer and mean it.

Question 24

What if a prompt is wrong for my job?

Accepted Answer

It sometimes will be. The tasks are evidenced, but the writing is generated, and generated writing has a failure rate. Treat any prompt as a first draft by someone who has read a great deal about your job and has never done it. You are the one who has done it.

Question 25

Why is my job missing, or thin?

Accepted Answer

A few honest reasons: Coverage follows the occupational record, and that record is better for some jobs than others. We cap each career at its best-evidenced tasks, so a broad job can look thinner than it is. New and hybrid roles lag the official taxonomies by years. It is growing — tell us what is missing and it goes on the list.

Find the work.
Then find the words.

What kind of work?

Not sure where you sit?

Pick the one that looks like your work

What the map lets you cross

Four heights of help

Execute

Improve

Decide

Become

The same work, seen by different jobs

Or find your job here

Or start from a subject

Or find your life, not your job

Can AI actually teach you this?

Or find it by who you are

Questions, answered

One role, across software

Find the work.Then find the words.