ONLINEAGENT_OPS 2026.Q3 HOME HUBSMODELSBUILDINGCOMPANIESFILMMAKINGSECURITYLIVING WITH ITSKILLSALL HUBS, ONE TERMINAL → ARTICLES CRAFT RECORD BLOG TOOLS SKILLS SEARCH /
HOME›ARTICLES›AI AT WORK: A BEGINNER’S GUIDE
ARTICLES · GUIDEBEGINNER

AI at Work: A Beginner’s Guide

What AI is good and bad at in your job, how to use it without trouble, and what the evidence on jobs really says, in plain words with sources.

READ12 min
WORDS2,301
SECTIONS9
TYPEGUIDE
CHECKED12 OCT 26

At work, treat AI as a fast intern who needs checking. It is quick at first drafts, summaries and rewording, unreliable on facts, and never the one who signs off. Used that way it saves time; trusted blindly, it gets people into trouble.

This page explains what it is good and bad at, how to stay out of trouble, and what the evidence says about jobs.

TL;DR — THE SHORT VERSION
  • Good for: first drafts, summaries of your own material, rewording, brainstorming and dull admin.
  • Not good for: facts you cannot check, decisions, anything confidential, and messages where it matters that you wrote them.
  • Staying out of trouble: know what may go in, say when it helped, and remember you own whatever you hand in.
  • Jobs: employers are naming AI in job cuts, but national figures barely move. The clearest effect so far is fewer first jobs.
  • Getting started: brief it like a new colleague, and check what matters.

What AI is actually good for at work

The plain point: AI is good at producing a first version quickly. It is much weaker at knowing whether that version is right.

Prefer paper? Print the one-page cheat sheet.

The comparison: a very fast intern on day one. They read in seconds and write fluently, but do not know your clients or rules, and will guess rather than admit it.

TaskWhy it worksHow to use it
First draftsemails, reports, postsWhy: a blank page is the slow part.Use it for: a starting point you rewrite in your own words.
Summariesnotes, long documentsWhy: you give it the material, so it has less to invent.Use it for: a quick overview, then read the parts that matter.
Rewordingshorter, plainer, politerWhy: it changes how things are said, not what.Use it for: tidying, then check no facts were added.
Ideasoptions, angles, questionsWhy: quantity is cheap, and you do the choosing.Use it for: a list to react to, not a decision.
Adminrotas, listings, routine emailsWhy: tedious, low-risk, easy to check.Use it for: the work nobody has time for.

Employers are already shifting this kind of work. In one survey of US employers, 38% said they had moved basic data processing from entry-level staff onto AI.1 Actual use still lags what the tools could do. Anthropic’s own research says AI “is far from reaching its theoretical capability” in the workplace.2

TAKEAWAY

Hand over the labour: drafting, tidying, summarising. Keep the judgement about whether it is right.

What it is not good for

The plain point: AI writes just as confidently when it is wrong, and it cannot take responsibility for anything.

  • Facts you cannot check. Names, figures, quotes and citations are where it invents things. Why does AI hallucinate? explains the reason.
  • Decisions. It does not know your situation, and it tends to agree with whoever is asking. How AI influences you back shows what was measured.
  • Confidential material. Once pasted, it cannot be taken back.
  • Messages where you matter, such as an apology or a reference.

The tool is not the problem; skipping the check is. A US judge who sanctioned two lawyers for citing invented cases wrote that “there is nothing inherently improper about using a reliable artificial intelligence tool for assistance”, but the lawyers still had to make sure their filings were accurate.3

Blaming the chatbot does not work for companies either. When Air Canada argued its website chatbot was a separate legal entity, a Canadian tribunal disagreed, as reported by a law firm.4

AI AT WORK · HAND OVER THE LABOUR, KEEP THE JUDGEMENT
A fast intern who needs checking, and never the one who signs off.
HAND OVER
Yes: First drafts of emails, reports and posts
Yes: Summaries of your own material
Yes: Rewording: shorter, plainer, politer
Yes: Ideas to react to, and dull admin
KEEP FOR YOURSELF
No: Facts you cannot check
No: Decisions
No: Anything confidential
No: Messages where it matters that you wrote them
Reasoning, 1 Oct 2026 — summarises this page’s sections on what AI is good for at work and what it is not good for.
◈ IN PLAIN TERMS

Before relying on an answer, ask: can I check this? And if it is wrong, who pays? If you cannot check it and you will be the one blamed, do the work yourself. When not to use AI lists nine cases.

How to use it without getting into trouble

The plain point: three rules cover most of it: what may go in, what you must say, and who answers for the result.

Rule 1: what may go in. The line is the account, not the tool. A personal account and a company account under a written agreement can have very different terms. Anthropic’s consumer terms, for example, do not cover its business products.5 Never paste customer data, passwords or anything under a confidentiality agreement into a personal account.

Rule 2: say when it helped. Disclose where a reader’s decision would change if they knew: client work, published work under your name, figures and research.

Rule 3: you own what you hand in. Using a tool does not pass the blame to the tool. If your name is on the figure, you check the figure.

The comparison: AI is a borrowed van. You are still the driver, and the speeding ticket comes to you.

A team AI policy you can actually adopt turns these rules into a one-page template. Using AI at work covers the same ground for one person, and what happens to what you type explains storage and training.

TAKEAWAY

Ask which tools and accounts your employer approves. Ten minutes of asking removes most of the risk.

Job applications

The plain point: using AI to write a CV or cover letter more clearly is fine. Letting it add a skill, a result or a year you do not have is lying.

The comparison: an editor makes your story clearer. Only you know if it is true, and the interview will ask.

  • Fine: tightening bullet points, matching real experience to the advert, practising interviews.
  • Not fine: invented numbers, tools you have never used, banned live help in interviews.
  • The useful instruction: “Make this shorter and clearer. Do not add anything I did not say.”

If you hire people, do not screen with AI detectors. In a peer-reviewed test, seven detectors wrongly flagged more than half of essays by non-native English writers as AI-written.6 Using AI on a job application has the full line and a checklist.

Making AI output usable for everyone

The plain point: AI output can look fine and still fail people who use screen readers, captions or high contrast.

The comparison: AI describes a photo like a stranger who never read the page: what is in it, not why it is there.

  • Alt text (the hidden description a screen reader reads aloud) should say why the image is on the page. A decorative image needs none.
  • Automatic captions need a check of names and technical terms, the words a viewer cannot guess.
  • Important words belong in real text, not inside a generated image.

Researchers also found that the prompt used to make an image does not work as its description.7 Is your AI output usable? gives a two-minute check.

What the evidence says about jobs

The plain point: AI is costing some people their jobs, but so far the effect is too small to show in national figures. The clearest sign is fewer first jobs.

Two things are true at once. Employers have named AI as a reason in announced US job cuts since 2023.8 Yet researchers at Yale find the mix of jobs Americans hold is not changing faster than before ChatGPT.9 An announcement is the employer’s stated reason, not a proven cause.

The comparison: a few houses knocked down on a long street. Terrible for those inside, yet the street looks the same from the end of the road.

The bottom rung is the thing to watch. The Dallas Fed found young workers in the most exposed jobs are being hired less, rather than laid off.10 The researchers note interest rates and post-pandemic cutbacks may also explain it.

The evidence points both ways. In one survey of US executives, far more expect AI to increase entry-level hiring in 2026 than to cut it.11

Be wary of “AI-proof jobs” lists. Some lean on a 2013 Oxford paper that estimated which jobs could be automated, while its authors said they were not predicting how many would be.12

Which jobs are safe from AI? gives the short answer first, then sorts jobs by how good the evidence is; AI and work covers both sides. The work AI cannot take is an argument, not a measurement.

TAKEAWAY

Judge the tasks in your own week, not your job title. The repeatable tasks with a clear right answer move first.

How to start

The plain point: start with a task you know well, so you can tell when the answer is wrong.

The comparison: brief it like a capable new colleague on day one. Say what you want, who it is for, what to avoid and what the answer should look like.

  1. Pick one dull, low-risk weekly task, such as meeting notes.
  2. Write a two-line brief: goal, audience, format.
  3. Edit the draft, do not restart. Say what to change: “Keep it, but shorter and less formal.”
  4. Check what matters: names, figures, dates and promises.
  5. Keep private material out unless an approved work account covers it.

How to write AI prompts and what makes a good prompt explain briefing step by step, with everyday examples. How to ask AI well covers habits for everyday work. Running a shop, salon or café? AI for a local business shows where it helps, and why you should never generate the product or a review.

Where to go next

Questions people ask

Can I paste confidential work information into an AI chatbot?

Not into a personal account. Never paste customer data, passwords or anything under a confidentiality agreement into one, because once pasted it cannot be taken back. A company account under a written agreement can have very different terms; Anthropic’s consumer terms, for example, do not cover its business products.5

Is it okay to use AI to write a CV?

Yes, to make it clearer. Letting it add a skill, a result or a year you do not have is lying. A useful instruction is: “Make this shorter and clearer. Do not add anything I did not say.”

Is AI really taking people’s jobs right now?

Some, but so far the effect is too small to show in national figures. Researchers at Yale find the mix of jobs Americans hold is not changing faster than before ChatGPT.9 The clearest sign is fewer first jobs: the Dallas Fed found young workers in the most exposed jobs are being hired less, rather than laid off.10

QUICK CHECK · 3 QUESTIONS

Pick an answer to see if you got it. Nothing is recorded.

1. How does the page suggest you think of AI at work?

  • As an expert who can sign off on decisions
  • As a fast intern whose work needs checking
  • As a search engine that is always correct
Show the answer

Answer: As a fast intern whose work needs checking. The page says AI is quick at drafts and summaries but unreliable on facts, and is never the one who signs off.

2. If AI helped with a report that has a wrong figure in it, who answers for it?

  • The company that made the AI tool
  • Nobody, because the mistake came from the AI
  • You, because you own whatever you hand in
Show the answer

Answer: You, because you own whatever you hand in. The page says using a tool does not pass the blame to the tool: if your name is on the figure, you check the figure.

3. Which use of AI on a job application does the page call fine?

  • Making your real experience shorter and clearer
  • Adding a skill you plan to learn later
  • Rounding your results up to look more impressive
Show the answer

Answer: Making your real experience shorter and clearer. The page says making a CV clearer is fine, but adding a skill, result or year you do not have is lying.

SOURCES
  1. 1ZipRecruiter Economic Research, More Jobs, Higher Bar: The 2026 AI Employer Report, 29 Jul 2026, read at source 23 Sep 2026“38% of employers have shifted basic data processing away from entry-level workers and onto AI”. ↑
  2. 2Anthropic Research, Labor market impacts of AI, March 2026, read at source 10 Sep 2026AI “is far from reaching its theoretical capability”. ↑
  3. 3Mata v. Avianca, Inc., S.D.N.Y., Opinion and Order on Sanctions, 22 June 2023, read at source 16 Sep 2026“existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings.” ↑
  4. 4McCarthy Tétrault, Moffatt v. Air Canada, 19 Feb 2024, read at source 16 Sep 2026, quoting the tribunal“It makes no difference whether the information comes from a static page or a chatbot”. ↑
  5. 5Anthropic, Updates to Consumer Terms and Privacy Policy, 28 Aug 2025, read at source 16 Sep 2026the updates “do not apply to services under our Commercial Terms, including Claude for Work”. ↑athe updates “do not apply to services under our Commercial Terms, including Claude for Work”. ↑b
  6. 6Liang et al., GPT detectors are biased against non-native English writers, Patterns, July 2023, read at source 17 Sep 2026they “incorrectly labeled more than half of the TOEFL essays as ‘AI-generated’ (average false-positive rate: 61.3%)”. ↑
  7. 7Das et al., From Provenance to Aberrations, CHI 2024, read at source 11 Sep 2026“prompts for T2I models cannot be used as alt text alternatives”. ↑
  8. 8Challenger, Gray & Christmas, monthly Job Cut Announcement Report, 6 Aug 2026, read at source 23 Sep 2026“Since 2023, when AI was first tracked as a distinct reason, it has been cited in 184,538 job cut announcements.” ↑
  9. 9The Budget Lab at Yale, Tracking the Impact of AI on the Labor Market, read at source 9 Sep 2026“more recent changes do not seem any more pronounced”. ↑a“more recent changes do not seem any more pronounced”. ↑b
  10. 10Federal Reserve Bank of Dallas, Young workers’ employment drops in occupations with high AI exposure, Jan 2026, read at source 9 Sep 2026the fall “is not due to layoffs but to a low job finding rate for young workers entering the labor force”. ↑athe fall “is not due to layoffs but to a low job finding rate for young workers entering the labor force”. ↑b
  11. 11Strada Institute for the Future of Work, Entry-Level Hiring in the AI Era, May 2026, read at source 17 Sep 2026“Nearly three times (2.7 times) as many senior talent leaders expect AI use to increase entry-level hiring in 2026 as to decrease it”. ↑
  12. 12Frey & Osborne, The Future of Employment, 17 Sep 2013, read at source 23 Sep 2026“We make no attempt to estimate the number of jobs that will actually be automated”. ↑

ZipRecruiter Economic Research, 2026 AI Employer Report, read 23 Sep 2026 · Anthropic, Labor market impacts of AI, read 10 Sep 2026, and consumer terms update, read 16 Sep 2026 · Mata v. Avianca and McCarthy Tétrault on Moffatt v. Air Canada, both read 16 Sep 2026.

Liang et al., detector bias study, read 17 Sep 2026 · Das et al., CHI 2024, read 11 Sep 2026.

Challenger, Gray & Christmas, 6 Aug 2026 report, read 23 Sep 2026 · Yale Budget Lab, AI labour tracker, and Dallas Fed, young workers study, both read 9 Sep 2026 · Strada, entry-level survey, read 17 Sep 2026 · Frey & Osborne, 2013 paper, read 23 Sep 2026.

Read dates are those recorded on this site’s linked detail pages. The comparisons on this page (fast intern, borrowed van, editor, stranger describing a photo, houses on a long street, new colleague) are this site’s own ways of explaining, not claims from these sources.

Free printable AI cheat sheets, and AI news every week.

Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.

SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK
aipromptgeneer://$ about this siteAI PROMPT GENEERAI explained in plain English. Every claim links to its source, with the date it was checked.NEWSLETTERWhat changed in AI each week, in plain English.SUBSCRIBE FREE ↗AI PROMPT GENEER · MACHINE ASSISTANCE DISCLOSEDEVERY ARTICLE CARRIES ITS OWN CHECKED DATE