The observations drawn on below were made by Matthew Wensing (VP Product & Design, customer.io) in conversation with host Aakash Gupta on The Growth Podcast, "How a VP of Product Uses Claude Without Producing Slop" (YouTube, published June 9, 2026). Full citation at the foot of this piece.
Here's a use case people are quietly betting on and shouldn't be: pointing AI at the problem of getting a room of executives to agree. Feed it the context, ask for a strategy doc or an all-hands narrative, and let the fluency carry the meeting. It's seductive because the artifact comes out polished, and polish reads like authority. It's also bound to fail. Not "needs a human in the loop" fail — structurally fail, for a reason that won't be patched in the next model.
The reason is simple to state. Executives are the best slop-detectors alive. Filtering noise is most of what senior judgment is. The whole job, by the time you reach the room that matters, is reading a document and knowing within a paragraph whether the person who wrote it understands the problem or is performing understanding. AI is very good at performing understanding. That is precisely the thing the room is built to catch.
Alignment isn't agreement. It's earned conviction.
Start with the definition, because people skip it and then wonder why the meeting didn't take. Alignment is not a document everyone nodded at. It's a shared conviction that survives contact with the next hard decision. Some cultures get there by "disagree and commit." Others demand shared consciousness on every detail. Either way, the thing being built is belief, and belief is downstream of whether the audience trusts that someone wrestled the problem to the ground.
A generated artifact can manufacture the appearance of that wrestling without any of it having happened. This is the trap. The doc is coherent, the structure is clean, the transitions are smooth — and none of that is evidence of thinking, because the model produces all of it whether or not the thinking occurred. A senior reader knows this in their body. They've spent twenty years learning to distinguish a conclusion that was earned from one that was merely formatted. You cannot format your way past that instinct. You can only trigger it.
The failure has a different face at every altitude
What makes this dangerous is that the failure doesn't announce itself. Wensing's observation is the useful one here: the symptom changes depending on where you sit, and most of the symptoms look survivable.
If you're senior and you push generated alignment into the room, you get the smile and the nod — "I guess we can work with this, but have you considered..." — assuming people feel safe enough to even say that. If they don't, you get the rolled eyes you never see and the quiet decision to ignore you, which is the worst outcome because it's invisible. If you're a director or a VP, lower in the stack, you get something blunter: the work gets filtered as noise and nobody feels any obligation to incorporate it. You worked all weekend on a point of view, and it gets no airtime, and you conclude the audience missed it. They didn't miss it. They processed it and discarded it in the first thirty seconds, exactly as they're trained to.
In every case the artifact "worked" in the narrow sense — it was delivered, it was on-brand, it read well. And in every case the only thing it was for, conviction, didn't happen. A deck that looks more aligned-with is not the same as a room that is aligned. The better the surface, the faster a senior reader files it as something to route around.
Why the next model won't fix this
The optimistic move is to assume this is a capability gap that closes on schedule. It isn't, and it's worth being specific about why, because the reason points at what the human still owns.
Generating alignment requires reading the room — knowing this audience just came out of an all-hands, just shipped the biggest launch in company history, is exhausted or exhilarated, and that a term you'd use casually carries three years of baggage and a couple of attached careers. Wensing had to stop a model from putting his cute internal names — unicorn, horse — into a leadership doc, because the room would fixate on the jargon and resist. The model didn't know that. It can't, yet: it has the words but not the social and political history that tells you which words detonate. That's not a hallucination in the factual sense. It's a misread of the people, and it's the more expensive error, because it doesn't look like an error until the room goes cold.
This is the part of the PM's craft that survives. Not generating the artifact — calibrating it. Knowing what a specific set of humans will trust, fixate on, resist, or quietly file away.
The actual mechanism: decomposition, not generation
So if you can't generate alignment, what produces it? The same thing that always did: a problem decomposed well enough that the room can see you understand its real shape. Wensing frames the whole AI-for-leaders question as a decomposition test — AI will happily solve a problem, but it flattens the problem space if you don't break it apart first. A flat, oversimplified projection of a hard problem is exactly what reads as slop, because the senior reader's first question is does this understand why we haven't already solved this? — and a flattened version visibly doesn't.
Alignment is won upstream, before any document exists, by doing the decomposition yourself: exploding the nasty problem into its real pieces, sitting with the parts that don't reconcile, and only then choosing what to put in front of people. The model is a phenomenal partner for that thinking — for pivoting raw material into shape, for stress-testing a mental model against reality. It is a terrible substitute for having done it. The conviction the room is looking for is the residue of the decomposition. There's no way to synthesize the residue without the work.
What to do now
Use AI to build the source — the cleanest possible model of the problem — and never to manufacture the consensus. If you're reaching for it to write the alignment doc, you've already inverted the tool: you're asking it to produce the one output it can only fake.
Before anything goes to a senior room, run the decomposition by hand and ask whether the document is evidence of thinking that happened or a performance of thinking that didn't. If you can't tell, the room can. And watch your own symptoms honestly — the polite "have you considered," the work that gets no airtime, the strategy doc that never enters the company's vocabulary. Those aren't communication problems to fix with a better template. They're the sound of generated alignment being filtered out in real time, which is the one thing these tools reliably produce: a more fluent way to be ignored.
Source. Matthew Wensing (VP Product & Design, customer.io), interviewed by Aakash Gupta on The Growth Podcast, "How a VP of Product Uses Claude Without Producing Slop," YouTube, published June 9, 2026: youtube.com/watch?v=yDeFGKaSoX8. All characterizations of Wensing's remarks are drawn from that episode; quotations are transcribed from it and lightly punctuated for readability. I have no relationship with customer.io, its employees, or the podcast.