Most of us have now had the conversation. Your manager mentions, sometimes in passing, sometimes in a town hall, that AI use is now part of how you do your job. There might be a specific tool you're meant to be using, an output target, or just a vibe — a sense that not engaging with AI isn't engaging with the future.

It's now common. An HRTech Edge survey of 1,295 business leaders found 58% of US companies require employees to use AI tools, with the remainder mostly encouraging it. The mandate is arriving in workplaces faster than the rules around it are being thought through.

What nobody tells you in the all-hands meeting is that most of these mandates arrive without a manual. There's an instruction but rarely a clear definition — what counts as 'using AI', what's off limits, how the results will be judged. You're being asked to incorporate something into your work without a brief.

That gap is where the real choice lives. Two responses are possible. You can adopt AI reflexively, on whatever task is in front of you, with whatever tool the company licenses, and slowly become more replaceable. Or you can adopt it strategically — picking the work where it genuinely makes you better and protecting the rest. The second response is the one that makes you harder to replace, not easier.

The mandate without the manual

When an employer says "use AI more," they almost never mean the same thing twice. The instruction can carry three quite different meanings: a demand for higher output because AI is supposed to make us faster, a cultural signal so the company looks modern, or a genuine shift in how the work gets done.

These call for different responses. If the goal is more output, you need to know which output and by how much. If it's a culture signal, you can probably comply by visibly using a tool the leadership team likes and getting on with your real work. The genuine workflow change is the most demanding scenario, and the one most worth investing in.

Asking your manager which of these is actually meant is one of the most useful conversations you can have. While you're at it, get specific on the rest of the operating brief: which AI tools are approved, what data you can and can't feed them, who reviews the output, and whether AI use is optional for the work that involves judgement. Most managers haven't thought about any of this clearly, and helping them get specific is a service to both of you.

Start with what bores you, not what you love

The advice everyone gives is to "find ways AI can help you." That's right but incomplete. A more useful framing: which parts of your job do you find tedious?

Research published in Management Science found that AI is most valuable for people who understand their own abilities and limitations. AI doesn't make average people great so much as it amplifies people who already know their own strengths and weaknesses. They know which tasks they're slow at, which they're inconsistent on, and which they secretly hate. Those are the tasks AI should take first.

For some people that's email triage. For others it's the formatting work that comes at the end of a project. For others still it's the spreadsheet grunt-work or the meeting notes nobody wants to write up. Take an honest look at where your time goes. The hours you'd happily delegate to a competent assistant, if you had one, are where AI earns its keep.

This is also where most people are using the tools badly. As I wrote about in the AI literacy gap, the average user is asking AI to summarise documents and tidy emails when it's capable of doing far more. If you've been told to use AI more, ask whether you're using it well at all.

Protect the parts that make you you

A study published in the British Journal of Educational Technology found that people who use AI more frequently report reduced interest and motivation in their topics. They also feel more dependent on the tools. The cost of over-using AI runs deeper than productivity: the thing you used to enjoy becomes the thing the AI does, and you become the person who clicks the button.

The fix is to be deliberate about what you protect. If you love writing, don't outsource the writing — use AI on the bits around it like the research, the formatting, the bullet-point summary nobody enjoys producing. If your strength is data analysis, let AI tidy the dataset and then think for yourself. The work that involves judgement — coaching, hiring, big calls — should stay with you, with AI handling the admin around it.

Being intentional with the parts of the work you'd be sad to lose is what this is about. Most of the AI-at-work conversation skips it, but the people I talk to sense it anyway. You don't have to hand over the bits you love.

Own everything that goes out with your name on it

The rule that will save your career is simple. AI output isn't a defence. When something goes wrong — a client report with a fabricated statistic, a contract with the wrong figures, an email to the wrong person with sensitive information attached — "the AI did it" is a sentence that won't protect you. The accountability sits with whoever signed off and sent it.

That means a few practical habits. Check the output, especially anything with numbers, names, dates, or specific claims; AI can hallucinate, which is the polite way of saying it makes things up confidently. Read it for tone, so the work sounds like you, not like a generic mid-Atlantic content writer. And know your organisation's policies. If you're using a personal AI tool, you almost certainly shouldn't be feeding it the company's confidential information. Use approved systems, or check what's allowed before you paste anything in.

The best AI users I know treat the output as a first draft, not a finished piece. That instinct alone will save you a hundred small problems. I'm not sure how reliably the habit survives a busy Tuesday afternoon — the data on this is thin, and I've watched plenty of careful people get sloppy under deadline. The people who keep at it tend to come out ahead anyway.

The two-track future

Zoom out from any single mandate and a pattern emerges. The wave of "use AI at work" is creating two kinds of workers, and the difference between them isn't enthusiasm, age, or technical skill.

Reflexive adopters use AI because they were told to, on whatever task is in front of them, with whatever tool the company licenses. Their output goes up a bit. Their judgement goes down a bit. Over time they slide into the replaceable category, because what they do is what the AI does — and the AI never costs more next year.

Strategic adopters take the mandate as a prompt to get clearer about what they're actually for. They figure out which parts of their work AI does well and where they themselves add value the AI can't. They get harder to replace, because they've made themselves into something the company can't replicate by issuing a Copilot licence.

The difference often shows up in small choices. Two analysts working on the same dashboard land in different places. One uses ChatGPT to write SQL queries, pastes the results in, and ships it — until the day a hallucinated metric goes out and the boss notices. The other uses AI to scaffold the queries, then reads them, runs them, and sense-checks the output against last quarter's numbers. Slower the first time. Steadier afterwards. Twelve months on, one of them is still doing the work and the other has been quietly absorbed into the tool itself.

What separates these two paths is intentionality. Talent and attitude don't predict the difference. Strategic adopters started with a question: where does this make me genuinely better, and where would it hollow out the work I care about? Reflexive adopters didn't ask. They just used.

It's also worth saying that the responsibility runs both ways. Employers who issue AI mandates without clear thinking — about what they're actually optimising for, and about where the human judgement still matters — get the workforce they deserve. As I've argued before, AI won't fix bad management; it will expose it. Workers who adopt strategically tend to expose those gaps faster than workers who don't.

If you're reading this with a mandate sitting on your desk, you've got more agency than the situation suggests. Don't just use AI because you've been told to — use it where it makes you better, and look after the rest. The people who get this right won't be the ones their employers replace. They'll be the ones their employers can't afford to lose.

Quick answers

Where should I start using AI at work — and where shouldn't I?

Start with the bits you find tedious: formatting, summarising, triaging. Research in Management Science suggests AI helps most when you know your own strengths and weaknesses, so an honest self-audit is the first step. Don't reach for AI on the parts of the job you love — over-use is associated with reduced motivation and higher dependency, per a 2024 study in the British Journal of Educational Technology. Ask your manager up front about approved tools, what counts as 'using AI' for the target, what data is off limits, and who reviews the output.

What if AI gets something wrong in work that goes out under my name?

You own it. "The AI did it" isn't a defence — always check facts, names, numbers, and tone before signing off, treat AI output as a first draft rather than a finished piece, and follow your organisation's policies on confidential data.