Attention Is All You Need

Everyone is asking what artificial intelligence is going to replace. Fewer people are asking what artificial intelligence reveals. To me that seems like the more important question.

Because if a machine suddenly starts producing essays, code, images, business plans, marketing copy, and strategic recommendations, the first reaction for most self-employed people is fear. Which is fair. A lot of us make our living producing those things. So when a machine starts producing outputs that look like the outputs of intelligent work, the obvious question becomes: what happens to me?

But I think there is a better question. What is intelligence, really? How did a machine begin producing behavior that looks intelligent enough to make intelligent people feel threatened?

…

In 2017, a group of Google researchers published a paper called Attention Is All You Need. At the time, the paper was about machine translation — how do you get a machine to take a sentence in one language and produce the equivalent in another?

That may sound narrow now, but translation is a profound intelligence problem. To translate well, you cannot just swap one word for another. You have to understand context. "Bank" means one thing beside "loan." It means another beside "river." Meaning does not live inside the word alone. It lives in relationship.

Before this paper, most language models processed text step by step. One word, then the next, then the next, carrying memory forward as they went. Other approaches scanned for nearby patterns — small windows of meaning. Both worked. Both had limits.

The Transformer made a different bet. Instead of marching through language sequentially, it allowed every part of the input to relate to every other part. It turned language from a line into a field. The Transformer did not become powerful because it memorized words in isolation. It became powerful because it learned to determine what mattered in relation to everything else. This word matters because of that word. This phrase changes because of that context. This distant part of the sentence matters right here.

This is attention in a more technical and useful sense: not focus in the motivational-poster sense, but weighted perception — the ability to determine what matters in context.

Output is the fruit

Most of the panic around AI comes from a mistaken definition of intelligence. We treat intelligence as output: the essay, the design, the code, the strategy, the finished product. So when AI produces the output, we assume it has replaced the intelligence.

But output is not intelligence. Output is the fruit. The root is attention.

The Transformer paper reveals something about the nature of intelligence: intelligence-like behavior can emerge when a system learns to attend to relationships in context. That does not mean the machine is wise. It does not mean human intelligence and machine intelligence are the same thing. But it does show us that useful intelligence depends on knowing what to attend to.

I learned this the unglamorous way. For years, I sold products by working backward from search. Before I wrote the page, I wanted to know what someone typed into Google at the moment the problem became real. I would look at forums, competing pages, half-answered questions, support threads, and the language people used before they knew what they needed. The pattern became hard to miss: the offer that sold was rarely the one with the cleverest copy. It was the one placed in right relationship to a problem people were already trying to solve.

Your business is running on a pre-Transformer model

Remember the two approaches the Transformer replaced? Recurrence — processing step by step, carrying memory forward in a fixed sequence. And convolutions — scanning for nearby patterns, small local windows. Both worked. Both had limits. And both describe exactly how most self-employed people have been taught to run their businesses.

Recurrence is the step-by-step playbook. Follow my 7-step framework. Do this, then this, then this. Build the funnel, write the sequence, launch the offer. The entire online business education industry runs on recurrence — someone else's sequence, handed to you as a recipe. The problem is not that the steps are wrong. The problem is that the sequence was built for someone else's context, someone else's market, someone else's timing. You are carrying their memory forward, not yours. So you follow the steps, and the results do not come, and you assume the problem is you. The problem may not be you. It may be the architecture. You are processing your business in a sequence that was never calibrated to your relationships.

Convolutions are what happens next. The playbook did not work, so you look around. You scan what people near you are doing. Your peer launched a course, so you launch a course. Someone in your space is doing webinars, so you try webinars. A competitor niched down, so you niche down. You are pattern-matching on a small local window — what is visible, what is nearby, what seems to be working for someone who looks like you. But you are reading the surface. You do not see their sales conversations, their client history, the ten failed attempts before the one that worked, the specific relationship between their offer and their market that made it land. You are copying the output without access to the intelligence underneath it.

Both of these approaches produce activity. Neither one produces intelligence. Because neither one asks the question the Transformer was built to ask: given everything in the input, what actually matters here?

That is the shift. Not a better step-by-step. Not a better peer to copy. A fundamentally different architecture — one where every part of your business is allowed to relate to every other part. In that architecture, a client conversation can inform your offer language, a failed launch can reshape your understanding of the buyer, a support request can reveal a product, and delivery friction can become content. Experience stops merely accumulating and starts compounding.

Most self-employed people are surrounded by signal. They just do not have a system for attending to it.

A business is a system of relationships

A business is not a product. It is not a brand or a funnel or a collection of tasks. A business is a system of relationships — between a person and a market, between an offer and a problem, between trust and risk, between language and recognition, between past work and future opportunity.

When those relationships are healthy, the business moves. When they are misread, everything feels harder than it should. The owner says "I have a marketing problem," but often the real issue is a broken relationship. The offer is not in right relationship with the market. The language is not in right relationship with the buyer's lived experience. The price is not in right relationship with the urgency of the problem. The founder is not in right relationship with their own authority.

I had a client once who was a gifted brand strategist — the kind of person who could walk into a company, spend a day listening, and hand them a positioning statement that made the CEO cry. But she could not sell her own work. She could name the hidden tension in a client's brand after one conversation, but when it came to her own business, she kept flattening the work into phrases like "brand clarity" and "strategic positioning." She had incredible intelligence about other people's businesses but struggled to turn that same perception on herself. That is not a marketing problem. That is an intelligence problem.

The missing Transformer

The Transformer changed AI because it gave machines a better way to attend to relationships. A self-employed business changes when the owner develops the same thing.

Most businesses have inputs — client work, sales calls, emails, questions, objections, support requests, ideas, proof, patterns. But without attention, those inputs stay scattered. The business has raw material. It does not have a Transformer.

In a business, the Transformer is not a tool. It is a practice of relating the parts. You take what happened in delivery and ask what it teaches marketing. You take what happened in sales and ask what it reveals about the offer. You take what people search and ask what it tells you about demand. You take what clients praise and ask what it says about value. The intelligence appears when these things stop living in separate rooms.

What does transformation actually look like? A project becomes a case study. A process becomes a framework. A repeated explanation becomes content. A client result becomes proof. A buyer objection becomes better positioning. A search pattern becomes a distribution strategy. A failed launch becomes a clearer offer map.

That is how experience compounds. Not automatically. Through attention.

What AI actually reveals

AI is not the end of self-employment. It is the end of unconscious self-employment.

If your business is built only on producing generic output — words, images, code, plans that anyone could produce about anything — then AI is a serious threat. But if your work depends on perceiving context, making distinctions, understanding relationships, and directing useful action from inside a specific world you actually know, then AI becomes something else entirely: a collaborator, a mirror, a force multiplier.

But only if you bring intelligence to the relationship. Because AI without business intelligence just accelerates confusion. It helps you produce more content without knowing what needs to be said. It helps you create more offers without understanding the buyer's moment. It helps you move faster in the wrong direction. It helps you generate options when what you actually need is discernment.

The danger of AI is not that it replaces intelligence, but that it exposes where intelligence was never developed.

A designer I know started using AI the way most people are told to use it: proposals, blog posts, captions, follow-up emails. Within a few weeks, she was producing more than ever. But the business felt exactly the same. The same leads. The same uncertainty. The same scramble to explain why her work mattered. AI had not changed the business. It had accelerated the existing pattern.

Attention is all you need

In AI, Attention Is All You Need meant something technical: a model architecture could be built entirely around attention mechanisms. But in business, in self-employment, in the practical question of how to remain useful and solvent in a world where machines can produce, the phrase carries a larger lesson.

Attention is not merely where you look. Attention is how intelligence forms. Because attention reveals relationship. Relationship creates meaning. Meaning guides action. And action, repeated in right relationship to reality, is what intelligence actually is.

So when self-employed people ask me how to survive the age of AI, I do not think the answer is to become more machine-like. Produce faster. Automate everything. Crank out more output.

The answer is to become more intelligent. More attentive. More capable of seeing the business as a living system of relationships that you — not the machine — are responsible for perceiving.

The machine became powerful by learning to attend. The self-employed survive by learning to attend better — not to words in a sentence, but to the relationships that make their work valuable.