Language
What form does communication take when intelligence becomes part of the medium through which meaning moves? A first-principles investigation of representation, attention, personal mediation and delegated action.
For thousands of years, language has been the interface between minds. We write, speak, draw, summarise, search, and explain because meaning has to be turned into a representation another person can receive.
AI changes that constraint. If a machine can transform, filter, expand, compress, verify, personalise, and eventually act on, meaning before another person encounters it, then language is no longer necessarily the entire communication process. It becomes one possible representation produced by a larger system.
This book does not argue that words disappear, or that AI captures complete human meaning. Language does not stop mattering. The question is what happens to communication when intelligence moves into the channel itself, and language becomes one representation among those a larger system can produce:
What form does communication take when intelligence becomes part of the medium through which meaning moves?
What the book builds
The book describes one imagined system, a layer that sits between people and AI systems as they communicate, and it grows that layer part by part. In its last chapter the layer is named for what it would be in practice: a modest OpenCode plugin and an experimental communication layer, with its final form still to be narrowed by implementation and experiment.
Part I: A New Language. Intent, concepts, representation, reconstruction. What happens when a source no longer has one fixed representation? The part begins with how a receiver rebuilds meaning from whatever a representation gives them, then follows fixed pages, previews, enhancement, source-linked semantic records, sidecars, typed relationships, preservation obligations and the representations that sit between human language and machine state. Together they make a Semantic Browser: the layer working on one source, choosing a representation for a task and checking what survived.
Part II: Too Much Information. Attention, selection, preservation, mediation. Once representations are cheap, what deserves to reach a person at all? Information units, task relevance, prior knowledge, standing semantic subscriptions, levels of resolution with a stop rule, and delivery timing compose into a Personal Information Radar: the same layer attending to a stream, with a decision record for every candidate and no single score.
Part III: The Personal Experience. Personal communication state and adaptation. When communication adapts to a person, what may the layer keep, who decides, and what must stay explicit? Declared, observed and inferred information kept apart, no stable person type, evidence turned into inspectable policy, session overlays that shed, user-authored friction, boundaries that travel between services, and beliefs the person can see, culminating in a personal layer that can always be switched off.
Part IV: The Second Reality. Prediction, agency, delegation, speech, discovery, ecosystem. What happens when the layer begins to predict, speak or act for someone? Prediction measured and kept apart from identity, what a person wants to keep, an explicit grant with an action gate and a data-use gate, speech typed by the act, private discovery in stages, and the new failures that appear at population scale. The book ends by asking where the word language stops being the right one.
Read in order, the parts follow the layer as it observes communication, keeps track of intent and concepts, preserves meaning, adapts its representation, maintains scoped personal state, predicts carefully, allocates agency, holds bounded authority, speaks and acts within it, and meets other people’s systems.
How the book reasons
Every transformation in this book, whether compression, expansion, personalisation, filtering or delegation, creates an obligation: to measure what survived it. And every mediation must remain inspectable and contestable by the person it serves. A third principle follows: a successful operation in one domain never automatically becomes evidence or permission in another. Similarity is not relationship. Preference is not performance. A model of a person is not the person. Prediction is not authority. Private computation does not legitimate its purpose.
Much of the book’s machinery comes from refusing to collapse concepts that look adjacent but are not interchangeable, and each separation is stated plainly.
What kind of claims this book makes
Three kinds of statement appear. Established background is what prior research supports independently of this book, stated with its populations and limits. Argument is what the book proposes: a mechanism, a distinction, a policy that prior work motivates without settling. Proposal is a design for a test that has not been run. This book reports no experiments of its own, and the people, sessions and figures in its examples are constructed. Where it describes a study, it is a design. The evidence the argument would need, and the ways it could be wrong, are recorded alongside the manuscript and are not presented as results. The thesis itself, that language moves up a stack and becomes one rendering of a deeper communicative process, is a hypothesis, and the last chapter says what would count against it.
Why first principles
Rather than starting from today’s chat interfaces, the book starts from the underlying communication problem: one mind holds state another does not have, representations are lossy, attention is finite, recipients differ, and acting for another person requires authority. Representation, selection, adaptation and delegation follow from those constraints, as do the obligations of preservation, provenance and contestability that bind every stage.
You do not need to believe AI will reshape language to use this framework. The distinctions between representation and retrieval, attention and comprehension, preference and benefit, prediction and authority already matter to systems being built now. The book ends at the horizon where language stops being the whole communication system, while remaining one of its most important rendered representations, and says plainly how far that claim has been tested, which is not far yet.
Chapters
The Interface Between Minds
Communication is a gap between what one mind holds and what another rebuilds. What happens to that gap when intelligence enters the channel?
What the Receiver Rebuilds
Receiving a representation is not receiving its meaning. What a person recovers depends on what they bring, and that is the standard any mediated representation must meet.
Communication Was Never Just Words
Different representational channels give a receiver different evidence to rebuild from. When does a second channel help, and when does it only decorate?
The Fixed Page
Why must the representation be fixed before the author knows the reader or the task? The fixed page has real virtues and a real limit.
Visual Compression
What happens when a source is compressed into a smaller representation meant to help a reader decide what deserves attention? Something becomes easier to see, and something may disappear.
Semantic Enhancement
Once a reader is inside a source, can the representation change locally to make what matters easier to see, without adding claims the source never made?
Meaning Becomes the Source
If a source can be rendered many ways, each rendering has to reinterpret it. What would it mean to keep one structured interpretation and generate the renderings from that?
Enhanced Interaction
What additional material may appear beside a source while someone is reading it, and how can the reader always tell what came from the source and what came from elsewhere?
Similarity Is Not Discovery
When a system puts one piece of information beside another, what relationship is it claiming exists between them? Similarity is only one relation, and it is not the one that usually matters.
What Survived the Transformation?
When a representation is transformed, how do we decide whether it preserved what mattered? Survival, loss, addition and gain, and why gain is not addition.
The Language Between Languages
Between ordinary human language and executable machine state there is a ladder of representations. What does each make easier, what does each give up, and what happens to a person's intent as a statement becomes easier for a machine to check?
The Semantic Browser
What the architecture of Part I looks like when it operates as one system: a source, an interpretation, declared obligations, a task, a chosen representation and a check on what survived.
Too Much Information
When a system can produce more potentially useful communication than a person can attend to, the problem stops being how to represent information and becomes which of it should reach them, when, and in what amount.
A Video Is Not an Hour
Before a system can decide what deserves attention, it needs units smaller than the source. What counts as a unit, and why does the answer depend on both the source and the task?
Important to Whom?
Once a system has candidate information units, what makes one of them matter for the task at hand? Relevance is a relation between a unit and a declared purpose, and it is constrained by what a faithful account of the source requires.
Don't Tell Me What I Already Know
Given the same source and the same task, what would actually add something for this particular person? A narrow model of what someone knows, with the evidence behind it, and the rule that known does not mean safe to omit.
From Channels to Semantic Channels
What would it mean to subscribe not to a creator, a feed or a topic but to a relationship in the world? A standing, inspectable relational interest over changing information, and the care it needs.
Compression Has Levels
Once an item deserves attention, how much of it should reach the person first, and where must compression stop? Resolution as a ladder, with a stop rule based on what has to survive and not on length.
When Should the AI Speak?
Once several items qualify for attention, when should each one be surfaced? Delivery as a decision separate from selection and resolution, balancing the cost of delay against the cost of interrupting, using only context the person declares or the system can observe.
The Personal Information Radar
What does the whole attention architecture look like when information is continuously competing to reach one person? One afternoon in the Radar, with a decision record in place of a score, inspectable coverage, and a hard boundary at what the system cannot see.
The Personal Experience
When a channel starts to remember its relationship with a person, what may it keep, who decides, and what must stay explicit? The first step is to make the person a declared input to how things are shown, inside limits the person cannot trade away by accident.
There Is No Universal Best Interface
If a system is going to remember how to adapt to someone, perhaps it should discover what type of user they are. The evidence does not justify that shortcut, and what replaces it is more useful: separate sources of adaptation, each scoped and each with its own authority.
From Observation to Policy
What a system may deliberately remember across interactions, how remembered evidence changes its standing, and how it can alter communication without becoming a theory of the person.
The Person Changes During the Day
The communication requirement changes faster than the person does. Standing policy can persist, but the conditions of this moment should shape a session-scoped overlay that disappears, and becoming a rule should always be a question.
Sometimes the Best Interface Adds Friction
What should a system do when the easiest next action conflicts with the person's own declared objective? Friction as a contract the person wrote, with a visible source, an easy override and a clear expiry, and never a judgement the system makes for her.
Protection That Travels
If a person has declared a communication boundary once, how can it be carried into another service without carrying the history that produced it? A portable boundary, a boring adapter, and a receipt that says honestly what survived.
What the System Believes About You Must Be Visible
A system that carries interpretations of a person, and not only what she said, holds beliefs that change what she sees. A belief must show where it came from and what it changes, and being visible is not the same as being authorised.
The Personal AI
What happens when every mechanism of Part III affects one communication at the same moment? One session, six pieces of information, separate namespaces that stay separate, and the question of whether effective communication needs a personal model at all.
How Much of You Can an AI Reproduce?
How much of a person can a system predict well enough to improve communication, and how would we know? Recall, estimation, prediction and reproduction are different claims, and prediction is evidence about behaviour, not evidence that the system has become the person.
A Model of Me Is Not Me
A model may resemble a person without being her, and may predict her without being authorised to act for her. Resemblance is not identity, and authority is granted, not inferred from predictive accuracy.
What Do I Want to Keep?
Before asking what authority a system has, ask who does which part of an activity. Capability, desire and permission are different questions, and what a person wants to keep belongs to the activity, not to the person as a trait.
From Conversation to Delegation
Agency allocation says what a person wants the system to do. Delegation says what it is actually authorised to do: an explicit grant, an action gate, a separate data-use gate, narrowing authority, receipts, and revocation that dominates.
What May My Proxy Say for Me?
A message sent in a person's name is an action, and not all speech acts carry the same consequence. Speech authority is typed by the act, speaking for someone is not speaking as them, and provenance should explain agency, not merely generation.
Private Discovery Between People
How can a system help two people discover a relevant connection without revealing more private information than the discovery itself requires? Discovery before disclosure, a match is not permission to reveal, and one person's consent does not authorise another's information.
A Society of Proxies
A pairwise-safe interaction does not imply a population-safe system. Local authorisation is not global safety, many agents are not a society, emergence is not legitimacy, and permission to send is not a right to be heard.
The Final Extent of Language
What has Language become by the end of the book, and at what point does 'language' stop being the right description? Words as one rendering of a deeper communicative process, the boundary where other systems begin, and what would count against the thesis.