The vocabulary gap

Originally published on billfaruki.substack.com on March 21, 2026.

There's something I've noticed about the people who are genuinely good at using AI — not the people who post screenshots of ChatGPT writing their emails, but the people who've internalized inference-based reasoning as a cognitive tool.

They all sound like philosophers.

I know because I do it. I catch myself in conversations with other practitioners, and we're three levels deep into abstraction before either of us realizes we've left the building. We're using metaphors stacked on metaphors. We're gesturing at something real, but the words keep sliding off it.

The reason proficient AI users default to abstract, philosophical language when discussing AI is that we literally don't have the vocabulary for what we're experiencing.

Modern inference AI is roughly three years old in any meaningful public sense. That's not enough time for a culture to develop the language infrastructure needed to describe a fundamentally new category of human experience.

Think about what happens when you sit down with a capable model and do real work. The kind where you're iterating on a complex problem, where the model pushes back on your assumptions, where you adjust your prompting mid-stream because you can feel the reasoning drifting.

What is that? It's not programming. It's not searching. It's not having a conversation. It's not delegation. It is something genuinely new — a form of directed cognitive partnership that has no name.

When the automobile first appeared, people called it a “horseless carriage.” Photography borrowed the entire language of painting. The internet got “surfing,” “pages,” “folders” — metaphors from the physical world because the digital world didn't have language yet.

AI is in the horseless-carriage phase right now. The people who are furthest ahead in actually using it are the ones who feel the vocabulary gap most acutely.

The thing where you can feel a model's reasoning quality shift mid-response. There's a texture to it. No word for this.

The skill of structuring a prompt so the model's inference path is shaped before it begins generating. This isn't “prompt engineering” — that term has been diluted to meaninglessness. It has no name.

The moment when a model's output genuinely changes your thinking. Not because it told you a fact, but because the structure of its reasoning revealed an angle. It has no name.

The learned intuition for which tasks are inference-shaped and which aren't. Arguably the most important skill in professional AI usage. It has no name.

It slows adoption. When the best users can't articulate what they're doing in concrete terms, training others becomes incredibly difficult.

It distorts the public conversation. Practitioners defaulting to philosophical abstraction reinforce the narrative that AI is mysterious, unknowable. This feeds both the utopian hype and the existential panic.

It advantages whoever fills the gap. The people who develop the vocabulary will have outsized influence on how the entire field develops. Language doesn't just describe reality — it shapes what we can think.

The solution isn't to stop being philosophical. The abstractions are doing real work — they're the best tools we have for communicating about genuinely novel experiences.

But we should be conscious of what's happening. When you find yourself reaching for metaphor after metaphor in an AI conversation, recognize it for what it is: evidence that you've encountered something real that language hasn't caught up to yet. That's not a failure of your thinking — it's a feature of being early.

The vocabulary will get built. The question is whether it gets built deliberately, by people who understand what they're naming, or whether it accretes accidentally from marketing decks and Twitter threads. I know which one I'd bet on producing better outcomes.