How to work well with AI (it's not about the prompts)

Here’s the strange state of AI in most companies right now: everyone has the tools, everyone uses the tools, however almost nobody was shown how to work with them — and the internet’s answer to that gap is another list of magic prompts. (You know the play: leave a comment with a heart emoji and I will send you the 100 magic AI prompts. Magic!!)

We’d like to offer something more honest than that. Less sexy, though. Working well with AI is not a prompting trick. It’s four core habits, all of which you already (hopefully…) use with people, plus one frame around everything that keeps you out of trouble. Here they are, with examples.

Habit one: brief it like a person you respect

Watch what usually happens. Someone types “write a LinkedIn post about our new maintenance service” and gets back something that could be from any other company on earth. Verdict: the tool is overrated. Case closed. Back to the basics.

Now run the same task the way you’d hand it to a good employee:

“This is for operations managers in manufacturing who already have a maintenance contract elsewhere. After reading, they should question whether their current provider is really checking what matters. Use our inspection checklist (see attached) as the hook, keep our usual tone (direct, no superlatives, not super formal), and don’t promise response times — we never promise response or execution times in public.”

Same tool, same person typing. Maybe three minutes more effort. The second briefing produces a usable draft; the first produces mush. The mush was never in the machine. It was in the brief. A few minutes of actual briefing replaces twenty minutes of disappointed re-prompting, and unlike prompt tricks, briefing is a skill your best people already have. They just haven’t aimed it at the machine yet, because nobody told them how to do it.

Habit two: treat the first answer as draft zero, always

(That goes for AI and colleagues alike.) The second mistake looks like using AI correctly: prompt in, answer out, done. A vending machine that doesn’t even give you this blissfully cold Coke.

The people getting real value work in rounds. Take a customer complaint about a late delivery:

• Round one: “Draft a reply; here’s the complaint and what actually happened.”
• Round two: “Too apologetic — we did partially deliver on time; acknowledge the miss, but state that plainly.”
• Round three: “Now read it as the customer. What would still annoy you about the response, and where are any risks of this reply?” The machine finds its own weak spot, oddly enough, quite reliably.
• Round four: a human makes the final call and hits send.

That’s four minutes of dialogue instead of one minute of vending, and the difference in output is not subtle. Draft zero is where the work starts, not where it ends. You’ve read plenty of shipped draft zeros this week. You recognized every one.

Habit three: feed it your company

The model has read half the internet and exactly nothing about you. That asymmetry explains most disappointing AI output: your prices, your tone, your standard clauses, how you handled this very same customer last March all live in your systems, invisible to the machine at the moment of asking. So the machine addresses your long-time customer and friend as “Dear esteemed client”. Whoops.

The fix is unspectacular. A sales team that collects its ten best offers in one place, writes its tone rules down once, and maintains a short “we never promise X” list turns generic drafts into drafts that sound like the company. A few afternoons of collecting what you already know.

And here, deliberately mid-article and not in a dedicated compliance box at the end, is where the legal frame enters — because feeding the machine is exactly where it becomes real and, to a certain amount, risky. What you give an AI tool is a data decision. Customer names, health details, salary data, anything a contract or the GDPR (or in Switzerland, the revDSG) protects: that material needs a sanctioned tool with the right guarantees, or it stays out. The practical form of this is not a policy nobody reads. It’s a two-list rule everyone can recite: what never goes into a prompt, and which tools are approved for the rest. Companies that skip this step don’t move faster; they just move blind, in tools they’ve never vetted. Working well with AI and working responsibly with it are the same discipline. The sloppy version isn’t quicker. It’s just uninsured.

Habit four: place it in the process, don’t sprinkle it on top

AI is an accelerant, not a repair kit. If your offers take two weeks because four people approve them sequentially and nobody remembers why, a faster draft simply waits two weeks with nicer margins. You’ve automated the part that wasn’t the problem.

So before the tools: map the process. Where does time actually go, which steps exist by decision and which by habit. Then place AI where it pays, typically at the start (drafts, research, preparation) and the end (checking, summarizing), with humans holding the middle, where the decisions live. That placement is also where responsibility gets simple: for anything that touches a customer, a contract, or a person’s data or livelihood, a named human reviews before it leaves the building. Since 2025, EU rules even expect companies to ensure their people understand the AI they use; less scary than it sounds, since reading pieces like this one is literally what that duty looks like in practice. The point was never bureaucracy. The point is that someone, with a name, stands behind what goes out. That’s not an AI rule. That’s just how serious companies have always worked.

The closing habit: decide where the saved hour goes

Do all four things above and AI will hand your team time back. Here’s the test almost nobody applies: at the end of the day, where did that hour go? In many companies it evaporates into more small stuff, same day, higher pace. In the good ones, someone decided: the salesperson calls one more customer, the project lead thinks a decision through instead of postponing it. Same hour, entirely different value. That decision is a leadership call, it takes one meeting, and it’s the difference between AI as a hamster-wheel accelerator and AI as an actual advantage.

None of this needs a transformation programme. It needs a briefing habit, a review habit, a few afternoons of collecting, one process map, two lists, and one decision about the saved hour. Most of it is learnable in a workshop; all of it goes faster with someone who has done it before, in a company like yours.

Just ask.