Back in 2024, my team at Blacfox was building a campaign for a client in enterprise software.
We were under time pressure. The client wanted positioning work, and wanted it fast. Positioning only holds up when it is grounded in what the market actually says. The honest way to get that is to go and ask. Interview buyers. Read the transcripts. Write down the words real people use for the problem they are trying to solve. It is thorough work.
And under the pressure, a member of my team decided to speed it up. He asked an AI to review the market’s feedback for us. The model swept up everything public about the category. Analyst reports. Vendor whitepapers. Blog posts. Review sites. It stitched all of it into a clean, well-structured summary of what the market wanted. It was articulate. It was balanced. It read like the kind of brief you would pay a research firm a small fortune to produce.
We built the campaign’s messaging on it with both hands.
The campaign failed. Completely and undeniably.
The client, understandably, wanted to know what had gone wrong. So we did what we should have done at the start. We went back to actual buyers. We called them. We listened.
What we heard sounded almost nothing like what the AI had told us.
The summary said the market cared about seamless integration and total cost of ownership. Crisp. Logical. Exactly what a sophisticated buyer is supposed to say. Except the buyers were not saying it. The vendors were. The machine had remixed the sellers’ story and presented it as the buyers’ reality.
The real buyers talked about something else entirely. They were terrified of data migration. They talked about the political risk of championing a platform switch that might fail in public. They did not trust their own teams to actually adopt a new tool. Emotional. Specific. Messy. And worth the entire engagement.
None of it appeared anywhere in what the machine had read, because buyers do not publish their fears. They say them in private, to someone safe who asks. Some of these buyers had never said them out loud at all until we asked.
We had built our campaign on a hallucination. Not the AI’s hallucination. Ours. We had hallucinated that a synthesis of public content was the same thing as market intelligence. It was a well-structured fiction, and we paid full price for the difference.
I am telling this story now because that fiction has since become an industry.
In August, researchers from Harvard and MIT published something called MatrAIx. It is a set of 8.3 billion AI personas, roughly one for every person alive. The idea is that you can test your product, your app or your agent on this synthetic population instead of on real people. Around it, builders are wiring up listening systems for machines. There are protocols so agents can hear each other. There are preference databases, so your agent can look up what “you” want without ever interrupting you.
Piece by piece, the machinery of commerce is learning to consult a copy of humanity instead of humanity itself.
I did not want to argue with that from an armchair. So I have been field testing it aggressively, across a sample of 500 humans in the United States. That research deserves its own articles, and it will get them later in this series.
In my last essay I argued that human desire is the backbone of commerce, and that machines know nothing of desire. Hold that thought, because it is about to earn its keep.
Call that copy a proxy: a stand-in that speaks for a person, so you never have to talk to the person. It is a tempting trade. And the pitch has real charm. Who still listens to actual people? That is so 1999.
For desire that is already settled, honestly, take the deal. Let the fridge guess the milk. Let the agents renew the toner. A want that is already formed, already spoken and already settled is exactly what should be automated.
But a proxy of a person is not the person. It is an interpretation. It is somebody’s retelling of what the desire is really about. Here, the somebody is a model. Listen to the machine version of your buyer and you are standing one step away, at least, from the human whose desire is the reason you are in business.
At one step away, appeasing that buyer becomes a game of chance. Appeasing means giving people what they want, and you are betting on somebody else’s account of what they meant. My agency ran that bet once, at full stake. The retelling we trusted was fluent about everything except the three fears that decided the deal. Those fears were the part it could not carry, because a fear is a feeling.
And there is something bigger here than appeasing. Shaping desire is the work of commerce. It is where the money lives. Nobody buys the big, considered thing because they woke up wanting it. Somebody understands the buyer well enough that what gets said resonates, and the desire moves.
That is how the trade actually works. You influence someone when what you say resonates with them. What you say resonates when you truly understand them, all the way under the want. And that understanding is exactly what a proxy can never give you. Resonance does not travel through an interpreter. There is no influence at arm’s length. You cannot shape a desire you cannot feel.
Spoken want is the raw material of all good marketing. Gather enough of what a market says it wants and you can make powerful inferences about how to position to that market. That is the heart of my method. This essay stands on it.
The proxy’s gap is in the interpretation. Under every want sits a root: the emotion feeding it. My buyers said they wanted integration and total cost of ownership, exactly as every vendor summary said they would. Underneath those words sat three roots. Fear of a botched migration. Fear of championing a failure in public. Fear of a team that would not adopt.
Two buyers can speak the same want for different reasons. The reason is where the deal is decided, because hearts and minds change at the level of feeling.
A root gets confided to someone safe. It gets understood by someone who has carried something like it. It does not survive being averaged.
A proxy can hand you the want. It cannot hand you the wanting.
There is a second problem. It is real, though the case never rests on it.
A market’s desire moves like a river. Every proxy is also a photograph of that river. A census of desire starts going out of date the moment it is printed, because appeasing the wants it recorded is exactly what creates the wants it never caught. Ice cream begot gelato.
A better model compresses the past faster. It still cannot know a sentence before a human speaks it. Worth remembering, and never the foundation. Even a machine that caught every sentence in your market the moment it was spoken would still be standing outside the root. And the root is where the decision is made.
Follow the logic one step further and it turns commercial. When everyone runs the same models, everyone’s machine holds the same words. Intelligence stops being an advantage and becomes a utility, like electricity.
What you cannot buy off a shelf is the felt understanding of what your buyers’ wants grow from, because that understanding only forms between humans in real contact. Your competitors can generate everything you can generate. They cannot feel your market for you.
That understanding is the last proprietary asset in the agentic economy, and it belongs to whoever is close enough to feel it.
Before you decide I am against the machines, look at the other half of my ledger, because I delegate to AI with both hands.
It drafts my emails and my proposals. It formats my documents. It generates variations. It tightens my language. It clips and summarises what I produce. It does all of it brilliantly.
The method I teach uses AI deliberately. It processes the interview data you provide. It generates positioning from that data. It builds discovery guides. It turns real market intelligence into messaging. I did not build a methodology and then reluctantly bolt AI onto it. I built one around what AI is magnificent at. And at midnight it drafts code beside me, while I build the Power Listening app as its only developer.
The line I hold is inputs versus outputs. Here is the policy, and you are welcome to copy it word for word.
AI drafts, summarises, transcribes and clips what you produce. It never reads the market for you, and it never invents feedback.
Ally, never proxy. Outputs, always. Inputs, never.
This is the chain I have argued for years, and the proxies change nothing about it except its value. Income comes from value. Value comes from understanding. Understanding comes from listening.
Value is what it is worth to appease a desire, and the shaping of that desire belongs to whoever understands it best. That is why true human listening is the secret weapon of the agentic era. It is the only thing that produces the understanding the whole trade runs on.
So to know human desire, we must listen to humans. Not to machines. Not to the machine version of humans.
Here is the drill. It takes one meeting. Open the deck, the plan or the playbook you are running right now. For every claim it makes about your market, write two things in the margin. The name of the last real human you heard say it. The date you heard it.
No name, no date, and that claim is synthetic, whoever wrote it. It makes no difference whether it came out of a conference room or out of a model. Count the synthetic claims. That margin is your listening scorecard, and your income is already reflecting it.
The machines will run more of the world every year. Let them.
When your market speaks its wants, be the one close enough to feel the roots they grow from.
Make it rAIn, KG