What happens when machines are listening

Prompting as a model for the next level of communication in consulting. I recently had an argument with an AI agent.

A proper argument. With exclamation marks, capital letters and a sharpness of tone that I wouldn’t have allowed myself to use with a human being. The prompt hadn’t delivered the result I’d expected — even though it had previously been optimised by that very same AI. What followed was a low point in my communication skills, which I mention here not without a touch of self-deprecation. The machine, famously lacking a nervous system, didn’t flinch.

But it was precisely this moment that taught me more about communication than many a seminar in recent years. And this is the story of it.

The web we cannot see

In day-to-day project work, communication functions surprisingly well. Not because we are always particularly precise, but because we humans catch each other’s mistakes.

Anyone who has worked in consulting long enough knows this: clients and team members rarely say exactly what they mean. Statements are contextualised, gaps filled in silently, ambiguities corrected invisibly. Our brain is a high-performance pattern-recognition machine. Precision costs energy. Filling in the blanks is quicker. So we fill in the blanks. Automatically. Mostly unconsciously.

In a typical status meeting, a sentence like this comes up at the end: “It’s important to us that the result is technically sound and makes strategic sense.” No one bats an eyelid. No one asks for clarification. The project manager thinks of risks, the business analyst of technical completeness, the Scrum Master of deliverability. Everyone has a picture — just not necessarily the same one. And yet things go surprisingly well for a long time, because somewhere along the way, someone quietly corrects the course.

This invisible safety net of interpretation, experience and social goodwill is not a weakness in the organisation. It is often the organisation itself.

What happens when the net is gone

Then comes the machine. And with it, a new form of honesty.

AI doesn’t fill in the gaps. It doesn’t ask for clarification when something is ambiguous, but instead decides — quietly, consistently, without a guilty conscience — on one interpretation. What previously looked like confident brevity suddenly seems astonishingly thin. The sentence has remained the same. Only the net beneath it has vanished.

The obvious reflex is: “The machine doesn’t understand that.” That’s true. But it doesn’t go far enough.

Because the real break occurs earlier — even before anything is formulated at all. Prompting asks a question that, in day-to-day project work, is left unanswered surprisingly often: What do I actually want? Not vaguely, not implicitly, not in the sense of “it’ll work itself out” — but specifically enough to be able to delegate it.

In consulting, this form of clarity is sometimes mistaken for being overly fussy. Anyone who asks too precisely is quickly seen as difficult. Yet precise communication merely shifts the work: forwards, to the sender, rather than backwards, to the team. The silent extra work does not disappear as a result. It becomes visible. And thus, for the first time, negotiable.

A calculation that should cause unease

Communication can be viewed, in simple terms, as a chain of probabilities. Let’s assume: in seven out of ten cases, I clearly articulate what I mean. The message is conveyed accurately in eight out of ten cases. My counterpart understands it as I intended in six out of ten cases.

This results in:

70% × 80% × 60% = 33.6%.

In roughly one in three cases, exactly what was meant comes across. The rest is taken care of by interpretation, experience and context. Or to put it another way: work that someone does without having been explicitly asked to do so.

Communication studies tell us: the truth lies with the recipient. It is not what the sender intended to say that counts, but what arrives in the other person’s mind — filtered through experience, expectation, emotion and context. It doesn’t take disastrous communication to cause a misunderstanding. Three reasonably human moments are enough.

The internet catches the rest. Until it no longer does.

What machines could teach us — if we listen

This is where the real transfer lies. It’s not just AI that improves when we prompt more precisely. We do too.

Anyone who seriously starts formulating prompts inevitably develops a habit that is taken for granted in consulting but is often neglected in practice: thinking before speaking. Not structuring after speaking, not clarifying on request – but clarifying one’s own intention before a task leaves the room.

That sounds trivial. It isn’t.

Imagine you’re delegating a task to a new colleague. First day, no shared context, no unspoken understanding. What would you say to them? You’d state the goal, not just the method. You’d give them what they need to know — not everything you know. And you’d say how you’ll know when it’s been done well.

These three steps aren’t purely AI techniques. They’re communication hygiene — and they work just as well in a briefing, a task delegation or a client meeting.

The only difference is: when talking to people, you can manage without them. The safety net catches you. When talking to the machine, you can’t. And that’s precisely why it’s such an effective training ground.

The machine isn’t the problem. It is the first counterpart in everyday project work that doesn’t provide an interpersonal safety net — and thus shows us just how much weight others have been carrying up to now.

In almost every team, there are people who have been working between the lines for years. Who catch things, fill in the gaps, and quietly correct. Who rarely appear in project status updates and are never praised in retrospectives.

What is described here is not theory. It is project practice in action in some places. In our team at Germany’s largest insurance company, we have started to treat prompting as a tool for better project communication, not just for better AI results. By consistently using prompting in our projects, we are embedding principles of clear task delegation, precise goal formulation and explicit expectation setting. We also use our work with AI tools as a mirror and a training ground for this.

No magic. No paradigm shift. Just the logical extension of a question the machine has asked us: What do you actually mean — exactly?

The next level of communication in consulting does not begin with better tools. It begins with the admission that a language model without a nervous system is currently giving us some of the clearest feedback we have received in years: unclear communication remains work. Only, it is often done by someone else.

The exclamation marks, by the way, were mine.