AI Literacy · Field Notes
AI Literacy in the Workplace
Everyone has access to AI now. That was meant to be the great equaliser. Like the invention of the pistol in the wild west. What a large collective seems blissfully unaware of is that a tool which multiplies whatever you give it will just as happily multiply incompetence - and I have sat in meetings watching people present data sets as their own, with no idea how badly it had exposed them. This is the version of AI literacy I keep giving my own team. Less about the tools. More about the judgement needed before you even touch it.
Too long to read now? Email us for a copy →
or write to hello@fraserpractice.com
The Stuff Landing on My Desk
I have received documents, emails and drafts lately of a quality that is genuinely hard to describe. Not just rough - wrong. Inaccurate, padded, confidently meaningless. And the people sending them had the audacity to present them as finished work: emailed over, or worse, opened in front of me in a meeting and walked through line by line as if they were useful. Wasting my time, the room's time, and - though they could not see it - making themselves look like an idiot in the process.
Here is the part none of them had clocked. AI did not make them look competent. It did the opposite. It took whatever judgement they had, or did not have, and amplified it for everyone in the room to see.
That is the whole point of the article, really. The rest is detail.
A Multiplier Does Not Pick Sides
The single most useful thing I have observed about artificial intelligence is this: the value of it is still limited by the intelligence applying it. Its ceiling is your ceiling - and so is its floor. Although the jury is still out on when the robots will finally take over…
It is a force multiplier, not a substitute for competence. And a multiplier is indifferent - it does not know good from bad, it will amplify excellence just as enthusiastically as mediocrity. Feed it competence and it is extraordinary; you will produce better work, faster, than you ever could alone. Feed it incompetence and it is a catastrophe with a fancy layout - poor decisions arrived at quicker, dressed in more convincing language, and far harder for anyone downstream to catch.
AI does not create competence. It exposes it.
So before any of the how-to, the honest starting point is an uncomfortable one: make sure the person at the keyboard actually knows what they are doing. There is no setting for that. No prompt fixes it. No subscription tier to manage it better.
And there is a quieter risk on the other side of the same coin. Lean on it too hard and you hollow yourself out. I have watched capable people lose the knack of drafting a simple email, or thinking a problem through, without reaching for it first. It is a fine line - augment yourself and you are sharper than ever; outsource yourself and you are finished the day the tool is down, throttled, or simply wrong.
Two Jobs, Two Different Bosses
Almost everything I use AI for falls into one of two buckets, and the difference between them matters more than any prompt you will ever write.
Category one - let AI lead. Standardised work. Anything that already exists in a thousand near-identical formats and only needs your specifics dropped in: NDAs, contractual terms and conditions, policies, procedures, templates, standard reports - the administrative workings every business runs on. These follow a recognisable path. You are inserting names, dates, project details and perhaps some financial details. For this, an AI first draft is perfect - fast, consistent, and usually more thorough than the version you would have rushed out yourself. Review it, obviously. But let it carry the load.
Category two - you lead, AI reviews. Anything creative, experience-based, personal, or that does not already exist in a dozen versions elsewhere. Articles. Strategy. A difficult conversation you are bracing for. A proposal that has to sound like you. This is where a lot of otherwise sensible advice gets it wrong.
The common thread is that AI should write the first draft and you refine on top. For this kind of work, I think that is exactly the wrong way round. Write a draft on top of an AI draft and the bones of it are still the machine's - the structure, the framing, the thinking underneath. You have edited AI, not written. The thinking has been outsourced, not accelerated, and you usually cannot feel the difference until someone asks you the one question you cannot answer.
So, I do it the other way. I write the first draft. Badly, sometimes, but it is mine. Then AI gets the second pass, and a deliberately narrow one. Check the spelling. Check the spacing and the wording. Tell me where the cadence drags and the read goes flat. Flag the weak arguments and the gaps. And then - this is the part that matters - suggest, do not change. Hand the suggestions back to me to accept or bin. The ideas stay mine. The voice stays mine. AI helps me say it better; it does not get to say it for me.
The more unique the work, the more human the thinking has to be.
The Split, at a Glance
Category 1 · AI Leads
Best for: standardised work where the format already exists.
Examples: NDAs, T&Cs, policies, templates, standard reports.
First draft: AI writes it.
AI's role: draft it; you review and sign off.
Flip it and you get: slow work, doing by hand what AI does well.
Category 2 · You Lead
Best for: creative, experience-based or personal work.
Examples: articles, strategy, proposals, hard conversations.
First draft: you write it.
AI's role: check, flag and suggest only; you decide.
Flip it and you get: competent, forgettable nothing - in your name.
Learn Prompting. Then Stop.
A lot of noise gets made about prompt engineering. Learn it at a basic level - enough to give context, state what you want, and specify the format you want it in - and then move on. It is nowhere near as important as the people selling courses on it would like you to believe.
Do one thing properly instead: set the back end up once - and schedule periodic reviews to ensure it still tracks where you want it to be - it could naturally progress as your experience of it does - but largely this should be done at the start and then ignored. Most tools have a memory or custom-instructions function sitting in the settings. Spend ten minutes telling it who you are, how you want things presented, the spelling you use, the tone you prefer. This prevents you from having to re-explain yourself at the start of every conversation.
After that, prompting is loose and forgiving. You can talk to it out loud, back and forth, like a person. You can type, paste, or dump in a half-formed mess and tidy it together - you can even voice note and chat to it. You can take an answer from one model and paste it into another to get a completely different angle on the same problem. There is no magic incantation. Clear thinking and comprehension wins every time - so spend your effort learning to think clearly, not learning to prompt.
The Limit You Will Actually Hit
A practical wall the beginner guides skip, and it becomes the real constraint the more fluent you get. It is not the prompting. It is the usage limit.
Once AI is genuinely woven into how you work, you start running into bandwidth and usage caps - the tier you are on, how much context the model can hold at once, how much you can get through in a day before it cuts you off - like a half-drunk reveller on a night out, just getting started, all too soon turfed out into the night five drinks too short of a good time. Get good enough and your own subscription becomes the ceiling: you cannot do the work to the standard you want, because you have run out of tokens, not ideas.
Treat AI as a toolkit, not a single app. Know your limits. Know which model is strongest for which job, and when to switch between them. Know what eats your context and what does not. It is an unglamorous, deeply practical part of the literacy required to produce large amounts of output - and it matters more than all this media about prompts.
What Not to Feed It
There is a line nobody draws clearly enough, so I will draw it here: be careful what you put in.
Most of these tools learn from, or at least process, what you feed them - and the free tiers especially are not a private vault. So treat the prompt box like a postcard, not a sealed envelope. Client data, commercially sensitive numbers, contracts under NDA, anyone's personal information, your own intellectual property - none of it belongs in a public model on a whim. The rule I give my team is blunt and it travels well: if you would not email it to a stranger, do not paste it into a chatbot.
This is not paranoia, it is basic hygiene. The competence the rest of this article is about includes knowing what stays on the inside. Plenty of otherwise sharp people have leaked something they should not have, because the tool felt private and personal. It is neither.
Use Your Own Intelligence
The line I keep repeating to my team is a simple one: the key with artificial intelligence is that you still have to apply your own. Every output gets reviewed - the facts, the numbers, the recommendations, the assumptions, the tone, the logic. All of it. Never confuse speed with quality. A document that took thirty seconds and reads beautifully can still be wrong in every way that counts, and polished writing is not evidence of good thinking. It is just polish.
The test is not whether AI helped. Of course it helped. The test is whether you understand what came out of it. Can you explain it? Can you defend it? Can you answer the question nobody in the room anticipated? Can you spot where it has quietly gone wrong? If not, you have not accelerated your understanding - you have replaced it. And that gap shows the moment the document is on the screen and someone starts to ask questions about it.
Because AI can write the document. It cannot attend the meeting for you.
Meetings have quietly become the great leveller. AI can build the slides, draft the report and polish every sentence. What it cannot do is sit in the room and explain why Recommendation Three beats Recommendation Two, hold the line on an assumption nobody thought to challenge, or read the Finance Director's face and adjust. That part is still yours, and it always will be.
You Can Usually Tell
You can spot it more often than people think, though the giveaways shift constantly. Today it is the over-formatting, the strangely tidy symmetry, the stylistic habits borrowed wholesale from whichever model is in fashion - those long dashes stitched through every other sentence among them. Tomorrow the tells will be different ones. Most of the imagery doing the rounds is generated too, and we are getting worse at noticing: one widely-cited study found people now correctly tell an AI image from a real photograph only about 38% of the time. Worse than a coin toss.
Quick test, since we are on the subject. Have a look at the image below and decide: real, or AI?
Real, or AI? Decide before you read on.
Made your call? Hopefully you clocked it as AI - a cat does not have the anatomy to flip anyone the bird, and no camera caught this. Clearly an AI production in this day and age, but perhaps the sort of thing that could have been passed off as genuine in yesteryear. There is an ongoing trend, though, with deepfakes and even food-delivery apps, where the image is doctored to make it look like the food is damaged - enabling the fraudster a free meal at the expense of the large conglomerate. Most often these costs are passed onto the restaurant, and it is becoming a huge issue at the ground level. The fakes are getting better and the tells are getting fainter.
The reliable giveaway is far simpler though: a person's grasp of their own material. Incompetence has signature moves - the confident hand-waving, the answer that evaporates under one follow-up, the inability to say why and not just what. You can see straight through the facade when someone has been augmenting themselves beyond their usual remit. Not because the writing is too neat. Because the understanding behind it is not there. And, funnily enough - the work they produced three months ago was basic, without depth and required multiple reworks, and all of a sudden it is ready for a boardroom in a Fortune 500.
Where to Actually Learn This
I have no intention of teaching any of the tools here - the web is drowning in that, and most of it is better than anything I would write. What is genuinely scarce is a shortlist that is not trying to sell you a two-thousand-dollar certificate. So here is where I would start. Pick one, finish it, and use it on something real the same week. That beats collecting ten more you never open.
Business and leadership
Worth knowing: Harvard Business School Online's AI Essentials for Business is a separate, paid programme (around $1,850), not the free tier. Do not conflate the two.
Prompting - keep it short, as promised
The model makers' own guides are the best, and they are free - OpenAI, Google and Anthropic all publish prompting documentation. Amazon's Foundations of Prompt Engineering on AWS Skill Builder is a solid free few hours if you want some structure. Read enough to deliver something beyond your usual limits, then stop.
Frequently Asked Questions
What is AI literacy in the workplace?
Knowing when to trust AI, when to challenge it, and when to ignore it. It is less about operating the tools - most are easy - and more about the judgement you apply to what they produce. Access is now universal; judgement is the part that still separates people.
Which AI tool is best for what?
There is no single winner, and they leapfrog each other constantly. The literate move is to treat them as a toolkit: learn the strengths of two or three, and switch depending on the task rather than staying loyal to one.
Is prompt engineering worth learning?
At a basic level, yes - enough to give context, state your goal, and specify the format. Beyond that it is oversold. Set your tool up once with good standing instructions, then spend your effort on thinking clearly. Clear thinking beats clever prompting every time.
How do I stop my team submitting poor AI-generated work?
Set the standard that every output is reviewed by the person whose name is on it, and that they must be able to explain and defend it. The problem is rarely the AI; it is people presenting work they do not understand. Make understanding the deliverable, not the document.
How much should I rely on AI at work?
Enough to multiply what you can already do, not so much that you cannot function without it. Use it to draft standardised work, to pressure-test and tidy your own thinking, and to move faster. Keep the judgement, the accountability and the original thinking human.
Use AI to multiply your capabilities, never to replace them. Every generation gets new tools - the ones who win are rarely those holding the newest, but those with the judgement to use them well. That has not changed, and it will not. Technology changes. Judgement does not.
Foundations - free, no coding
Elements of AI (University of Helsinki) - a platform-neutral, no-code grounding in what AI actually is and is not. Around 30 hours, and it costs nothing. →
The full field note
Want the whole thing as a PDF?
No forms, simply mail us and we'll send the full field note straight over. Tell us where you're stuck if you like, and we can start the conversation.
Email us for the PDF
or write to hello@fraserpractice.com