AI Can Produce Content. It Can't Decide What Deserves to Be Said.
The other night I asked a model to generate a full AI strategy for a 200-employee manufacturing company. Thirty seconds. Fourteen pages. Clean structure, right vocabulary, risks named, a prioritization matrix, even a governance section.
I read it twice looking for the crack. Honestly, there wasn’t an obvious one.
And if I had sent it to a real client, it would have accomplished nothing.
Not because it was bad. Because it answered a question nobody had bothered to ask properly.
AI collapses the cost of producing text. It does nothing to the cost of understanding, deciding, and standing behind what you put your name on. Those two things look alike when you’re staring at the deliverable. They have almost nothing in common.
A polished document is not a good answer
Models are excellent at turning an instruction into content. Summarizing, structuring, comparing options, adjusting tone to the audience. That’s real, it saves an enormous amount of time, and I use it every day.
But how well a document is written tells you nothing about whether it’s relevant.
A report can be flawlessly written and rest on a crooked reading of the problem. A deck can be persuasive and completely ignore the political dynamic that will kill it in committee. A strategy can contain every right word without proposing a single right choice.
Writing something that looks like a strategy is easy. Knowing which one is right for this organization, this year, with this leadership team, is a different job.
When a company calls to ask “how should we be using AI,” the mandate is almost never to hand over a list of tools. It’s to understand what they’re actually trying to accomplish, where it jams, which decisions are badly equipped, and what’s stopping them from moving.
And above all, what isn’t being said out loud.
They ask me for a conversational agent, and the real problem is sleeping inside a process nobody has documented in years. They want to automate a task the team hasn’t mastered yet. They think they’re short on technology when they’re short on clarity and alignment. I’ve written before that AI is a management turning point first, not a technology one: this is exactly what that means in practice.
AI can write the answer. It can’t decide the question was the wrong one.
OK… Mathieu, everybody says “humans still matter.” Got anything new?
Fair. It’s become a platitude. So let’s be specific about where humans still matter, because it isn’t everywhere, and it isn’t free.
What you never see in the deliverable
Inside organizations, the deliverable gives a misleading picture of the work.
You see the report, the roadmap, the deck presented to the leadership team. You don’t see the six interviews, the two disagreements, or the moment the VP of Operations let one sentence slip in a hallway and the whole analysis changed direction.
That’s where the value actually gets made.
With AI it’s even more visible. The tools move so fast that it feels like you can go straight from idea to automation. Spot a repetitive task, pick a platform, build an agent. On paper, perfectly rational.
In real life, the hard part is almost never building the agent.
It’s in the questions you have to settle first. Does this process deserve to be automated as-is, or are we about to pour concrete over a bad habit? Which decisions go to the system, which stay human, and at what point do we force an intervention? Where does the data come from, and what happens when it lies? Does the solution genuinely lighten the work, or does it just add a layer of complexity somebody will be maintaining eighteen months from now?
AI speeds up the analysis and the writing. It doesn’t make a single one of those questions go away.
Ghostwriting is the clearest case
You could assume a ghostwriter’s main contribution is writing on someone else’s behalf. If AI writes fast and well, their value should collapse.
Except that reduces the job to its visible part.
A good ghostwriter doesn’t start by writing. They start by listening and asking uncomfortable questions. They work to pull out the ideas someone holds but can’t quite articulate. They can tell the difference between an opinion picked up on LinkedIn last week and a conviction the person paid dearly for.
Writing is the last step. Not the main one.
AI can cover nearly everything else: transcribe the interview, cluster the themes, propose an outline, produce a very decent first draft. What it doesn’t do is notice the half-second hesitation before an answer, and understand that the hesitation was the article.
Same story in strategy consulting. The final document now takes a fraction of the time. The value moved upstream, to surfacing the right issue and framing it so a decision actually becomes possible.
When producing costs nothing, filtering becomes the job
There’s a second shift, and it gets discussed less.
As long as a report took two days to write, production capacity acted as a natural filter. Nobody wrote a forty-page analysis for fun. That filter just disappeared.
We’re entering a period where organizations will generate more analyses, recommendations, scenarios and summaries than they can reasonably read.
Yeah. We’re going to drown.
The problem is no longer access to information. It’s deciding which piece deserves attention, which analysis rests on solid assumptions, and which recommendation has earned the right to influence a real decision.
And watch the second-order effect: a poorly structured organization doesn’t get clearer with AI, it gets noisier, faster. More dashboards, more summaries, more recommendations, exactly the same capacity to act. I’ve seen this movie with enterprise software, and it rarely ends well.
A mature organization does the opposite: it uses AI to cut noise and concentrate human attention on the small number of files that genuinely require judgment.
What this changes for the rest of us
I’m not saying nothing moves. A big share of writing is going to be automated, and it already is: generic copy, first drafts, summaries, standardized communications.
Anyone whose value was mostly speed of production is going to take a hit. Seriously.
What gains value is harder to sell and harder to do:
- Writers: bring something beyond command of the language, because the language became a commodity.
- Ghostwriters: extract a line of thinking that doesn’t yet exist anywhere in written form.
- Consultants: get judged on the quality of decisions made, not the weight of deliverables filed.
- AI strategists: help the organization decide not what it can automate, but what it should, and under which conditions.
None of this is because AI can’t produce a sophisticated answer. It’s precisely because it produces them constantly, and somebody still has to decide which ones are useful, acceptable, and consistent with what the organization is trying to do.
The better the answers look, the more expertise it takes to evaluate them. Counterintuitive, but it’s what I keep seeing.
Where the line actually falls
The split won’t be between people who use AI and people who refuse to. The best ones already use it, and the holdouts will fold within two years.
It’ll be between people who use it to produce more, and people who use it to think better.
AI makes writing available to everyone. It makes nobody a writer, an advisor, or a strategist. It removes the difficulty of producing a document. It leaves untouched the difficulty of understanding an organization, carrying a recommendation, and deciding in the fog.
If you’re wondering what to automate this year, try the question backwards: what does nobody in your organization currently have time to properly understand? That’s probably where it’s worth starting.
Write to me if you want to talk it through. Have a good week!