We are building an internet for machines, not humans

When machines talk to machines, the only way for humans to cut through the noise is to say less, but mean more.

Abstract colorful tangled ribbon-like lines in orange, yellow, pink, and purple on a gray background, flowing rightward. Photo credit: Google DeepMind

There is one of the most fascinating shifts happening right now. We are transition from an internet built for human eyes to an internet built for machine comprehension.

Historically, humans wrote for other humans. Then came SEO (Search Engine Optimisation), and humans started writing for algorithms to reach humans. Now, we are entering a phase where AI models generate content that other AI models scrape, summarise, and synthesise.

This raises some massive practical and philosophical questions about the future of language, knowledge, and connection.

The new gatekeeper

Practically, the internet is splitting into two layers: the Front-End (what humans see) and the Data-Layer (what AI reads). We are building an internet where AI models are the primary creators and consumers of text.

Digital communication meant getting a human to click a link and read a page. Now, LLMs (Large Language Models) browse the web on behalf of users. If an AI reads your website, extracts the answer, and serves it to a user, the human never visits your site.

Structured data, clean formatting, and machine-readable content have become more important because they help systems extract information efficiently.

Close-up of a glossy black screen reflecting neon text: Perceive, Synthesis, Interfere, on a soft purple background. Photo credit: Google DeepMind

Language under pressure

Language evolved to do more than transfer facts.

People communicate through humour, ambiguity, metaphor, exaggeration, irony, and emotion. These elements are often inefficient from a computational perspective, yet they carry much of what makes communication memorable and meaningful.

As more content is optimised for machine interpretation, there is a risk that public writing begins to favour clarity at the expense of character. The incentives reward predictability, structure, and standardisation. Over time, distinctive voices can be crowded out by styles that are easier to process and reproduce.

This does not mean language will disappear. It means the conditions shaping language are changing. Writers may increasingly find themselves balancing two audiences: human readers and the systems that mediate access to them.

Woman in blue fog with red binary code projected across her face and body, creating a moody futuristic scene. Photo credit: cottonbro studio

Information and meaning are not the same thing

The more interesting questions emerge when we move beyond technology and consider what communication is actually for.

Information can be transmitted without understanding. A database can store facts. A machine can classify text. A system can summarise a book.

Meaning is harder to define.

When two people communicate, there is usually an attempt to establish a shared understanding of the world. Experience, context, interpretation, and emotion all contribute to that process. Communication is not simply the movement of information from one place to another; it is an attempt to connect perspectives.

Our information is filtered through layers of automated interpretation. Therefore, it becomes worth asking what survives that process. Are we receiving understanding, or are we receiving more and more refined representations of understanding?

The distinction may seem abstract, but it is injurious to how humans create culture, knowledge, and shared reality.

Glossy capsule with the text How do large language models work? on a purple patterned background, futuristic and abstract. Photo credit: Google DeepMind

The question of agency

There is also a question of who determines what information reaches us.

Every filtering system forms perception. Search engines did this. Social media algorithms did this. Systems that summarise, rank, and synthesise information do it as well.

The difference is that these systems are becoming gradually capable of determining not only what information appears, but how that information is framed before it reaches us.

Most people, as it appears, do not have the time to read thousands of pages on a subject. Summarisation is useful. Nevertheless, every summary is an interpretation. Every synthesis involves choices about what matters and what can be discarded.

The more we rely on mediated understanding, the more important it becomes to remain aware that someone—or something—is making those choices.

My opinion

As a writer, I don’t feel frustration or nostalgia, but looking at this structurally, communication should always serve the human end-state.

If you try to compete with AI on volume, speed, or explanation, you will lose. The machines will always out-produce and out-summarie you. AI-to-AI communication is an incredibly powerful utility. It can cure data overload. Having one AI summarise a mountain of medical research so another AI can cross-reference it with patient data is a massive win for humanity.

However, AI-to-AI communication should be the plumbing, not the architecture. If we stop writing for each other and only write for the models, we stop thinking for ourselves. The moment we view language purely as “data to be processed” rather than “experience to be shared,” we lose the very thing that made us want to communicate in the first place.