---
title: "Your AI Is Just an Alphabet"
slug: ai-and-alphabets
summary: "The frontier model you're paying for is available to millions of other people — access alone is not a durable advantage. What the 2,700-year history of a much older technology says about who actually wins."
tags: ai, llm, essay, prompt-engineering, creativity
ai_model: claude-fable-5
published_at: 2026-08-25T14:14:50.217+00:00
canonical_url: https://jarkkolietolahti.com/articles/ai-and-alphabets
markdown_url: https://cemecdwfdvdbvpvowuln.supabase.co/functions/v1/content-md?slug=ai-and-alphabets
---

I mean the title literally. The frontier model you're paying for is an alphabet: a shared raw material available to millions, carrying no lasting advantage in itself. Twenty-something letters, no waiting list — and almost nobody who owns them has written a book worth reading. That gap, between having the letters and having something to say with them, is the most useful way I've found to look at where we are with AI right now.

## Access alone is not an advantage

Take the alphabet: a, b, c, d, and so on down to whatever your language ends with (mine ends with å, ä, ö). Every writer working in a given language had the same set. Shakespeare didn't have premium letters. Neither did the person who wrote the instruction manual for my dishwasher.

Now open Claude, Codex, Copilot — whichever you use. Type something in. Your words get converted to tokens, the model proposes an answer, and you pay for the output. The same process happens for millions of other people who open the same chat window. Without your context, the model has no reason to care that it's you.

This is the part people seem to miss: **access to a frontier model, by itself, gives you no lasting advantage over anyone else — for the same reason that owning the alphabet gives no writer an advantage.** The raw material is rapidly becoming commoditized. It always was going to be — that's what makes it raw material. Most durable advantages — your data, your context, your tools and workflows — are wrapped around the model, not contained in mere access to it. That's ordering, not access.

## Default arrangement produces default work

If I rearrange letters randomly, I don't get a book. Not a good book, not a bad book — no book at all. The value is in the ordering, and the reason for that ordering comes from somewhere outside the alphabet: from knowing what a story is, what a reader needs, what came before.

The analogy has a limit worth naming: an LLM is not a passive alphabet. It can propose arrangements of its own. What it cannot decide is why a thing should exist, which constraints actually matter, or which result deserves to survive.

The same distinction holds one level up. If you throw undirected prompts at an AI chat, you are unlikely to get a killer app, a headless SaaS, an amazing song, or a breathtaking picture that differs meaningfully from what everyone else gets with similar prompts and the same model. You get the model's characteristic default output — much the same as anyone else typing much the same thing.

And yet that's a fair description of a lot of what's happening right now. People are racing, token-maxing, prompt-engineering, hurling prompts at models and hoping something sellable falls out. It's the equivalent of typing letters faster and expecting literature to emerge from typing speed.

## The skill was never the tool

Nobody would say "I know the alphabet" as a professional qualification. The alphabet is assumed; the skill is writing. Ordering the letters into a good book, good marketing copy, good source code — that's where all of the value lives, and none of it is in the letters themselves.

I expect the same sentence to sound obvious about AI within a few years: "I have access to an LLM" will sound exactly as impressive as "I know the alphabet." The differentiating skill is whatever you bring to the ordering — purpose, taste, domain knowledge, judgment about what's worth making, the ability to recognize when the output is wrong.

The model supplies candidate arrangements at unprecedented speed. You supply the reason for making them, the constraints they must satisfy, and the judgment that separates an answer from something worth keeping. It can write. It cannot decide what deserves to be written.

## We have the letters. The Odyssey hasn't been written yet.

Here's the part that I find genuinely encouraging rather than deflating.

The Greek alphabet is roughly 2,700 years old. Fairly early in its history, the orally transmitted *Odyssey* was fixed in writing — and that arrangement of letters is still generating value today: translations, novels, films, screenwriters mining it for structure. One good ordering of a free, shared raw material, still paying out after nearly three millennia.

With AI, we are at the stage where we've just got the letters. We're collectively doing the equivalent of scratching practice words into wax tablets — and, predictably, most of what's produced is practice-word quality. Perhaps the masterworks of this medium haven't been made yet. Perhaps some already exist and we're too close to recognize them. Either way, we probably don't yet know which forms will endure, the same way early writers couldn't have predicted the novel or the screenplay.

That's not a reason for hype, and it's not a reason for despair about AI slop either. It's just where we are on the timeline: very, very early. The letters are on the table, the same set in front of millions. What matters now — same as it mattered 2,700 years ago — is learning what deserves to be written.