Should I Buy It?
“It’s ugly” doesn’t work with AI. “Too geometric, lacks chromatic contrast, titles are too close to the edge” works. The difference between the two is explicitation — the skill that will separate AI users in the coming years.
Origin
Michael Polanyi, philosopher of science, articulated in Personal Knowledge (1958) the concept of tacit knowledge: “We know more than we can tell.” We can recognize a face among thousands without knowing how — that knowledge is tacit, unverbalized. Explicitation is the inverse: making explicit what was tacit.
Oussama Ammar applies this concept to AI in 2026:
“You need to learn to explain, to make explicit… explicitation becomes a new skill… a near-autistic competency where you see every detail and know how to talk to the machine.”
The Theory
The problem of subjective judgment
LLMs cannot process subjective judgments as inputs. “It’s ugly,” “it’s bad,” “improve this” — these instructions contain no exploitable signal for a language model. The AI is forced to fill the void with its own patterns, which don’t match what you wanted.
Wittgenstein and the limits of language
Ludwig Wittgenstein (Tractatus, 1921): “The limits of my language are the limits of my world.” With AI, the formulation is even more direct: the limits of your explicitation are the limits of what you can get. If you can’t say it, the AI can’t do it.
The monetizable skill
Prompt engineering research (2022-2024) shows that the precision of context provided to an LLM explains 60-80% of output quality variance. It’s not model power that makes the difference on comparable tasks — it’s instruction precision.
In Practice
Reformulation examples:
| Vague instruction | Explicit instruction |
|---|---|
| ”It’s ugly" | "Too geometric, lacks contrast, titles are too close to the edge" |
| "Improve the text" | "Shorten sentences over 25 words, replace jargon with common terms, active rather than passive voice" |
| "The tone is wrong" | "Too formal for a content creator audience, remove polite formulas, add a direct hook" |
| "That’s not what I wanted" | "I wanted 3 distinct options, not one, and each option must fit in 140 characters” |
The practical exercise:
Every time you want to write something vague to the AI, stop for 30 seconds and ask: “What criteria would allow me to objectively say this result is good?” Those criteria are your instruction.
Nuances and Limits
Explicitation can become a creativity blocker if applied too early. In an exploration phase, a vague prompt can generate unexpected directions — ambiguity is sometimes productive. Explicitation is most powerful when you know what you want, less so when you’re still searching.
There’s also a paradox: the people who struggle most with explicitation are often those with the richest tacit knowledge (experts, craftspeople). Making explicit knowledge you master intuitively is cognitively difficult — which is precisely why it’s a learnable skill.
Sources: Polanyi, M. (1958). Personal Knowledge. Routledge · Wittgenstein, L. (1921). Tractatus Logico-Philosophicus · Ammar, O. (2026). Podcast