Yes, tell your boss you use AI at work in most workplaces now, because Dmitry Kargaev's book Don't Replace Me argues that hiding it has become the risky move and the real risks are unchecked output and leaked data. Kargaev wrote this page and owns dontreplace.me; it draws on chapters 8, 15 and 16 of his second edition (2026).

Why hiding AI use is now the riskier move

Chapter 15, whose rule is "show the work, own the output", opens with Shopify CEO Tobi Lütke's AI memo, written in March 2025 and posted on 7 April 2025. The book quotes one line from it: "Reflexive AI usage is now a baseline expectation at Shopify." It adds that Shopify put AI questions into performance and peer reviews. The memo covers one company, not yours.

The chapter's warning also runs the other way: "Being careless with data is a much faster way to lose your job than AI ever will be."

Check the policy and your account settings first

Read the written AI policy before you say anything, and check whether your personal AI account trains on your chats. Chapter 15 puts both on a "before Friday" list, together with booking 15 minutes with your boss and bringing before and after times for one task. A script for asking your manager or IT which AI tools the company allows, and whether a written policy exists, appears in chapter 1.

On training, the book says business and enterprise plans are generally excluded by contract, while consumer defaults change often. Look at the setting today instead of trusting memory. Chapter 6 says to turn training off if your work is sensitive.

A Career Trend explainer from 18 September 2026 says whether a tool keeps your inputs for training depends on the product, the account type and the terms your employer has signed, and that IT or security can confirm it. It also splits the question in two: whether the tool is approved for the task, and whether the policy requires you to say you used it. It quotes Ran Xu, a director in Gartner's Risk & Audit Practice, saying information entered into a generative AI tool can become part of its training set.

The five-part script for telling your boss

Chapter 15 supplies a short script for the conversation. These are its parts, in the book's order, with the kind of thing that goes in each:

The book also gives a one-line answer for the hallway question "Did you just ChatGPT this?" A plain reply built on the same rule, show the work and own the output, would be: "Yes, for the first draft. I checked the numbers and rewrote the rest."

This came from a book.

Don't Replace Me

200+ pages. 24 chapters. The honest version of what AI means for your career, written by someone who actually builds this stuff.

Get the Book →

Check the output before it leaves your desk

Chapter 14 gives a seven-step verification checklist. Among its steps: open every source and find the sentence that supports the claim, redo the math in a spreadsheet, check quotes word for word, and read the piece as its recipient would. The book warns that asking the same model "are you sure?" feels like checking, but "it's the same guesswork running a second time."

The chapter opens with Mata v. Avianca. Judge P. Kevin Castel called six cited cases "bogus judicial decisions with bogus quotes and bogus internal citations" in May 2023, and on 22 June 2023 he imposed a $5,000 fine on lawyer Steven Schwartz, the signing lawyer and the firm.

Be useful, not loud

Chapter 16, rule 16, "be useful, not loud", says secrecy is no longer the play, but the loudest AI talker is rarely the person whose work changed. AI makes volume free, so quality is the only thing readers can tell apart. The book's advice is to let results arrive first and explain the method when someone asks.

The chapter cites the survey that coined "workslop", run by the Stanford Social Media Lab and BetterUp and published in Harvard Business Review in September 2025. It covered about 1,150 US desk workers, and about forty percent said they had received some in the past month. The answers are self-reported. Chapter 15 uses the word for the risk of sending unchecked output.

The chapter's exercise is a one-week quiet test. Pick one weekly deliverable and use AI to make it better, not bigger: draft it, cut it by a third, check the numbers and add the one sentence the reader needs. If your team discloses AI use, add one line saying so. Then tally replies and follow-up questions for two weeks.

What to say when AI output you sent was wrong

Chapter 8, rule 8, "be the one who answers for it", gives a word-for-word correction called the "it's on me" script. Its reasoning: "A model can't be fired, sued or embarrassed, and the client knows it." The chapter opens with Air Canada, which argued that its chatbot was "a separate legal entity that is responsible for its own actions"; a tribunal rejected that in February 2024.

The script has five parts, in this order:

  1. Name what was wrong.
  2. Give the correct version.
  3. Own it: say it is yours.
  4. Explain how you checked the rest.
  5. State what changes.

Mention the AI tool once, as a fact about how the draft was made, and never as the culprit.

What other advice says

Sachi Kittur, vice-president of human experience and innovation at the Human Resources Professionals Association, told The Globe and Mail "You should absolutely be open about it." She also says how much to disclose is a matter of degree and context, and that using AI to lightly edit an email is probably fine not to mention. CNN Business, on 30 December 2025, says to check your employer's AI policies first, verify anything you take from AI and be transparent with your boss. A Willa explainer from 29 September 2026 says to tell your boss in most cases and frame it around results. It cites an October 2025 Cornerstone OnDemand survey in which 57 percent of US workers said they were reluctant to tell their manager or colleagues about their AI use.

What is not known

No source here gives a rule that fits every workplace. Career Trend says the sources it cites do not add up to a universal disclosure rule, and Kittur's view shows the amount worth disclosing varies. The Shopify memo covers one company.

The surveys cited here count how many people hide AI use or receive workslop. Training defaults on personal accounts change, so this page gives no current settings. The author's own article on this question says that in roles where the process is auditable, such as lawyers, doctors and researchers, the conversation should probably happen with your compliance team before your boss.

Where to read the book

Don't Replace Me: A Survival Guide to the AI Apocalypse is on Amazon as a Kindle ebook at $2.99, a paperback at $14.99 (174 pages) and a hardcover at $24.99. Each chapter ends with one rule and one exercise, and the scripts above come from chapters 8, 15 and 16. The introduction is a free download at dontreplace.me. The book sells nothing after it: no course, no community, no paid next step.