Prompting at work: the patterns that do most of the job
Prompt engineering sounds like a specialty, but at work it comes down to a handful of repeatable habits. Here are the ones that do most of the job.
“Prompt engineering” is a great phrase for selling a course and a terrible description of what actually matters at work. It makes a simple skill sound like a specialty — something with secret incantations you have to buy. It is not. For getting real work done, good prompting is a handful of habits anyone can learn in an afternoon, and then a lot of practice on your own tasks.
Here is what actually moves the needle, without the mystique.
Why most prompts fall flat
The single most common mistake is treating an AI like a search engine. People type three words — “email to supplier” — and get something generic, because generic is all they asked for. A language model cannot read your mind, your context, or your standards. It fills the gaps you leave with the blandest plausible average of the internet.
The fix is not clever wording. It is giving the model what a competent colleague would need to do the job: what you want, who it is for, what you are working from, and what “good” looks like. Most of prompting is just refusing to be vague.
The patterns that do most of the job
You do not need fifty techniques. You need these, used together:
- Give it a role and a goal. “You are a bookkeeper writing to a client who is 30 days overdue. The goal is to get paid without damaging the relationship.” A role plus a goal anchors everything that follows.
- Hand it the raw material. Paste the actual email thread, the messy notes, the draft. The model is far better at transforming your real content than inventing generic content. The more of your own material you give it, the more it sounds like you and less like the internet.
- Show one example. If you have a piece you like — a past proposal, a tone you want to match — paste it and say “match this.” One good example beats a paragraph of adjectives.
- Name the format. “Three bullet points, under 20 words each.” “A table with these columns.” “A 150-word summary a busy exec would actually read.” Unspecified format is where most of the disappointment lives.
- Then iterate — out loud. The first output is a draft, not an answer. “Too formal, make it warmer.” “Cut the second paragraph.” “You invented that figure — remove it.” The people who get the most from AI are not the ones with the perfect first prompt. They are the ones who are quickest to correct a bad one.
That is it. Role, material, example, format, iterate. Ninety per cent of the value, no marketplace required.
What you can safely ignore
A whole industry has grown up selling complexity you do not need. You can skip:
- The 800-word mega-prompt templates. For most business tasks they are overkill, and worse, you cannot tell which part is doing the work when the output is wrong.
- Prompt marketplaces. A pack of “500 prompts for professionals” is a pack of someone else’s context, not yours. Your own three-line prompt about your own client beats it every time.
- Chasing the “perfect” prompt. Iterating twice is faster than engineering once. Stop polishing the question and start correcting the answer.
The skill that matters more than prompting
Here is the part the prompt-engineering hype leaves out: the highest-value skill is not writing the prompt. It is judging the output.
A model will hand you a confident, well-formatted, entirely wrong answer — an invented statistic, a misread clause, a plausible summary that missed the point. If you cannot catch that, better prompting just helps you produce polished mistakes faster. The professionals who genuinely get value from AI are the ones who treat every output as a draft from a fast, eager junior: useful, quick, and never to be trusted without a check.
That judgement only develops one way — on your own real work, with someone who can show you where the tool is strong and where it quietly lets you down. Which is exactly what our hands-on workshops are built around: not prompt theory, but building something that works, on your actual tasks, before you leave the room. And if you want it built into how a whole team works rather than a few enthusiasts, that is the Sprint.
Prompting is not the hard part. Knowing what to ask for, and whether the answer is any good — that is the part worth practising.