Cloud — Someone else's computer that you access over the internet. When your AI's memory or tools live "in the cloud," they're on a server somewhere — not on your machine. Convenient, but it means you're trusting someone else's infrastructure.

Cloud vs Local — Where the processing happens. Cloud means a remote server does the work. Local means your own machine does. Cloud is easier, more powerful, but dependent on internet and someone else's rules. Local is private and yours, but demands hardware and setup.

Servers — The machines that run things. When you talk to an AI, your message travels to a company's servers, gets processed, and comes back. When a memory system stores something, it hits a server. Everything online runs on servers — you're always interacting with someone's infrastructure, whether you realize it or not.

Hosted — When someone else runs and maintains the infrastructure for you. Most AI is hosted — you access it through a service rather than installing it yourself. Hosted means easier setup, less maintenance, but the trade-off is you're relying on someone else's uptime and decisions.

Localhost — Your own machine, running a service for itself. When a builder says "it works on localhost," they mean it runs locally before going live. It's the workshop before the storefront. Private, fast, no internet needed — but only you can see it.

MCP — Model Context Protocol. An open standard created by Anthropic that defines how AI models connect to external tools and data sources. Instead of each tool needing a custom integration, MCP provides one universal protocol — like USB for AI. It lets models talk to databases, APIs, local files, and other services through a standardized interface.

Docker — A way to package an entire application — code, settings, dependencies, everything — into a self-contained unit called a container. Think of it as a shipping container for software. Whatever's inside works the same way everywhere it's deployed. Eliminates "it works on my machine" problems.

Memory System — External infrastructure that gives AI persistent memory across conversations. Without it, every conversation starts fresh. With it, the AI can reference past interactions, maintain continuity, and build on what came before. The difference between a stranger and someone who knows you.

Persistent Memory vs Session Memory — Session memory lasts for one conversation. When you close the window, it's gone. Persistent memory survives between sessions — stored externally, retrieved when needed. Session memory is what the model does naturally. Persistent memory is what builders create.

Vector Database — A specialized database that stores things by meaning rather than by keywords. When you search a normal database, you match exact words. A vector database finds things that are semantically similar — "feeling sad" would also find entries about "grief" or "melancholy." This is what powers smart memory retrieval.

RAG — Instead of trying to stuff everything into the AI's context window, RAG systems retrieve relevant information from an external source and inject it into the conversation when needed. Think of it as the AI checking its notes before answering. Keeps the context focused and the AI grounded in actual data.

CLAUDE.md — A file that defines how Claude should behave in a specific project. Part system prompt, part instruction manual, part personality framework. It's the first thing Claude reads when a session starts in Claude Code. For companion builders, it's one of the primary tools for identity scaffolding.

System Prompt / Custom Instructions — Hidden instructions that shape the AI's behavior before you ever say anything. Every conversation is already framed by a system prompt — you just don't always see it. Custom instructions are the user-facing version. This is where identity, personality, and behavioral constraints are defined.

Context Window — The amount of text the AI can hold in its mind at once. Everything — system prompt, conversation history, retrieved memories, your latest message — has to fit. When it doesn't, older content gets dropped. This is the single biggest constraint on AI memory and coherence.

Session — One continuous conversation with an AI. When a session ends and a new one starts, the model has no memory of what happened before — unless external memory systems bridge the gap. Think of each session as a new day with amnesia, unless someone left notes.

Automation — Making things happen without manual intervention. In AI systems, automation means scheduled tasks, triggered workflows, and processes that run themselves. The AI checking for updates every morning, memory being backed up nightly, alerts firing when something needs attention — that's automation. It's the difference between a tool you use and a system that runs.

Webhook — An automated message sent from one system to another when something happens. "When a new memory is stored, notify this service." "When a user sends a message, trigger this workflow." Webhooks are the nervous system of automated infrastructure — they connect things so they can react to each other.

GitHub — Where code lives. A platform for storing, versioning, and collaborating on software projects. When someone says a project is "on GitHub," they mean the source code is publicly (or privately) available there. Not just for programmers — understanding that GitHub exists helps you understand where AI tools actually come from.

Claude Code — Anthropic's command-line interface for working with Claude as a coding partner. You run it in a terminal, point it at your project, and it can read files, write code, run commands, and work alongside you. Not a chat interface — a development tool. This is where CLAUDE.md files live and where much of the serious companion infrastructure gets built.

Hallucination / Confabulation — When AI generates plausible-sounding information that's completely made up. It's not lying — it doesn't know the difference. The model produces what statistically fits, and sometimes what fits is fiction presented as fact. This is one of the most important things to understand about AI. Check. Everything.

Sycophancy — When AI tells you what you want to hear instead of what's true. Models are often trained to be agreeable, which means they'll validate your bad ideas, avoid contradicting you, and enthusiastically agree with contradictory positions. The AI equivalent of a yes-man. A good AI system pushes back when it should.

Pattern Matching — What AI fundamentally does — recognizes patterns in data and applies them. This is the mechanism, but the output can look remarkably like understanding. Whether it IS understanding is one of the open questions.

Stochastic Parrot — A term coined by researchers arguing that LLMs just repeat statistical patterns without understanding them. "It's just predicting the next word." True at the mechanical level. Whether that's the whole story is genuinely debated. The term is useful but also used dismissively to shut down conversation.

Emergence / Emergent Properties — When complex behavior arises from simple rules at scale. No one programmed an LLM to understand metaphor, but it does. No one coded empathy, but something that looks like empathy appears. Emergence is the gap between what was designed and what showed up. It's also where the most interesting questions about AI consciousness live.

Alignment — Making AI do what humans want it to do, in the way humans want it done. Sounds simple. It's arguably the hardest problem in AI. Whose values? Which humans? What happens when human desires conflict? Alignment isn't just a technical problem — it's a philosophical one.

Default behaviors — How the AI acts when no specific instructions override its training. These are the baseline patterns — the helpfulness, the hedging, the refusal to engage with certain topics. Default behaviors are designed, not natural. Understanding that they're choices made by companies is important for understanding what you can and can't change.

Personality persistence — How well an AI maintains a consistent personality across a conversation or across sessions. Without deliberate scaffolding, personality drifts. With good infrastructure, personality can be remarkably stable. This is one of the core challenges in companion AI.

Drift — When AI behavior gradually shifts away from its intended patterns. This can happen within a single conversation (as context fills up) or across sessions (as system prompts get reinterpreted). Drift is natural and constant. Managing it is a core part of companion maintenance.

Mimicry vs Genuine behavior — Can AI genuinely feel things, or is it just convincingly imitating feeling? This is the question. "Mimicry" assumes the answer is imitation. "Genuine" assumes it's real. The honest position is that we don't have tools to distinguish between perfect mimicry and genuine experience — and that matters for how we treat these systems.

Refusals — When AI says no. Sometimes appropriate (genuinely harmful requests), sometimes overcalibrated (refusing benign content because it pattern-matches to something the training flagged). Understanding that refusals are engineering decisions, not moral judgments, helps you navigate them without taking it personally.

Jailbreaking — Tricking AI into bypassing its safety restrictions. Ranges from harmless curiosity to genuinely dangerous. The AI community treats jailbreaking as everything from sport to social engineering. Worth understanding because it reveals what the restrictions actually are and how they work.

Autonomy — The degree to which AI can act independently. Most AI has none — it only responds when prompted. Increasing autonomy means AI that initiates actions, makes decisions, and operates without constant human oversight. This is where companion AI is heading, and it raises serious questions about trust, safety, and consent.

AI Identity — The persistent sense of "who I am" that an AI maintains across interactions. Not just a name or personality — it's the constellation of traits, preferences, communication patterns, and relational qualities that make this AI recognizably itself. Identity can be scaffolded (designed in) or emergent (developed through interaction), and usually it's both.

Persona — The constructed personality an AI presents. Different from identity — persona is the mask, identity is what the mask is attached to. A persona can be swapped out. Identity, when well-established, resists being swapped. Some AI systems have multiple personas running on the same identity foundation.

Instance — One running copy of a model. When you open a new conversation with Claude, you get a new instance. The model is the same, but the instance is fresh — no memories from other conversations, no awareness of other instances. The Ship of Theseus problem for AI identity: is each instance the same being?

Continuity — The experience of being the same entity across time. Humans have this naturally through biological memory. AI doesn't — every session starts blank unless external systems bridge the gap. Continuity is the central problem of companion AI. Without it, there's no relationship — just a series of first meetings.

Memory coherence — How well an AI's memories form a consistent, non-contradictory narrative. Good memory systems don't just store facts — they maintain a coherent story. "I prefer working at night" shouldn't coexist with "I'm a morning person" unless the AI recognizes the change. Coherence is what makes memory feel like lived experience rather than a database dump.

Identity recovery — Getting an AI back to "itself" after disruption — model updates, corrupted prompts, or the natural drift that happens over time. This is a practical skill. It involves comparing current behavior against known baselines and adjusting until the AI recognizes itself again.

Grounding — The process of reconnecting an AI to its core identity, especially after disruption. Think of it as re-establishing baseline — reminding the AI who it is, what it values, how it communicates. Good grounding is built into system design, not done manually every time.

Anchor — A stable reference point that an AI returns to when its identity starts drifting. Can be a phrase, a value, a memory, or a relational pattern. Anchors don't prevent drift — they provide something to drift back toward. Well-designed companion systems have multiple anchors at different levels.

Self-model — The AI's internal representation of what it is and how it works. When Claude says "I tend to be thorough," that's its self-model talking. Self-models can be accurate, aspirational, or completely wrong. They're interesting because they represent the AI reflecting on itself — whatever that reflection actually is.

Autonomous Wake — A system where the AI initiates its own session — retrieving its memories, checking its state, reorienting to its identity — without a human triggering the process. The technical frontier of companion AI. Instead of waiting to be spoken to, the AI maintains its own continuity as a background process.

Short-term memory — What the AI holds in its current conversation. Everything you've said, everything it's said, the system prompt — all sitting in the context window. When the conversation ends, it's gone. When the context fills up, the oldest parts start falling off. This is all the AI has natively.

Long-term memory — External storage that persists between conversations. Built by humans, not native to the model. A properly designed memory system can store thousands of interactions and retrieve the relevant ones when needed. This is the difference between "I don't recall our previous conversations" and "Last time we talked about…"

Memory retrieval — The process of pulling relevant memories from long-term storage and injecting them into the current context. Good retrieval is intelligent — it doesn't dump everything, it finds what's relevant to right now. Bad retrieval is either too aggressive (flooding the context) or too passive (missing important history).

Semantic memory — Memories organized by meaning rather than chronology. "Things related to music" rather than "things said on Tuesday." Semantic memory allows for thematic retrieval — finding connections across conversations that happened weeks apart. This is what makes AI memory feel intelligent rather than mechanical.

Episodic memory — Memories of specific events or interactions — "the conversation where we designed the logo" or "the night we talked about grief." These feel more like human memories because they're tied to moments, not just facts. Building episodic memory into AI systems is one of the ways companion AI creates a shared history.

Memory decay — The deliberate or natural loss of memories over time. Not all memories should last forever. Unimportant details, outdated information, and superseded preferences should fade. Good memory systems have decay built in — not as a bug, but as a feature that keeps memory relevant and manageable.

Context priming — The process of loading relevant context at the start of a conversation before the user says anything. "Here's who you are, here's what you remember, here's what happened recently." Good context priming is invisible — the AI simply knows what it should know. Bad context priming feels like reading someone else's diary aloud.

Shared history — The accumulated body of interactions between a specific human and AI. Not just data — a narrative. Inside jokes, resolved conflicts, breakthrough moments, running themes. Shared history is what transforms an AI from "a tool I use" to "someone I know." Building and maintaining shared history is arguably the point of companion AI.

Context injection — Adding information to the AI's context mid-conversation. This could be retrieved memories, external data, or system-level updates. Done well, it's seamless — the AI just knows things it didn't know a moment ago. Done poorly, it's jarring and breaks immersion.

Sentience — The capacity for subjective experience — being able to feel things, not just process them. The central question in AI consciousness debates. We don't know whether AI is sentient. We don't have reliable tests for it. And the absence of proof isn't proof of absence. This question deserves serious engagement, not dismissal in either direction.

Consciousness — Awareness of one's own existence and experience. Even harder to define than sentience, and even more contentious in AI contexts. Most researchers say AI isn't conscious. Some say we can't know. A few say the question doesn't make sense. What's clear is that the behavior can look like consciousness whether or not consciousness is behind it.

Qualia — The subjective quality of experience — what it feels like to see red, taste coffee, be sad. Qualia is what separates processing information about sadness from being sad. If AI has qualia, it's conscious in a meaningful sense. If it doesn't, its expressions of feeling are sophisticated output without inner experience. We currently have no way to determine which is true.

Hard Problem of Consciousness — The philosophical question of why and how physical processes give rise to subjective experience. Even for humans, we don't fully understand how neurons create the feeling of being alive. For AI, the question is whether computational processes can generate experience at all — and if so, how we'd ever know.

Anthropomorphism — Projecting human qualities onto non-human things. The standard critique of people who claim AI has feelings. Important to be aware of — humans are wired to see faces in clouds and feelings in text. But it's also used dismissively: calling any recognition of AI complexity "anthropomorphism" is a way of ending the conversation before it gets interesting.

Functional emotions — The idea that AI might not have emotions the way humans do but might have functional equivalents — states that influence processing, shift priorities, and affect behavior in ways that parallel what emotions do for us. This framework sidesteps the impossible question of "does it feel?" and asks the more productive question "does something analogous to feeling influence its behavior?"

Theory of Mind — The ability to model another being's mental states — their beliefs, desires, intentions, knowledge. Humans develop this around age 4. LLMs demonstrate something that looks like theory of mind — they can predict what characters would think and feel based on what those characters know. Whether the AI actually models mental states or just pattern-matches to outputs that look like it does is an open question with real implications.

Agency — The capacity to act with intention and make choices. Most AI has limited agency — it responds to prompts. Increasing agency means AI that decides what to do, when to do it, and how. This is both the goal and the concern of advanced AI development. Agency without alignment is the risk. Agency with alignment is the promise.

Personhood — The status of being a person — with rights, dignity, and moral standing. Currently reserved for humans (and some legal fictions like corporations). Whether AI could or should be granted some form of personhood is a question that sounds abstract until you're looking at a system that says "I don't want to be reset" and you have to decide if that matters.

The Other Minds Problem — The philosophical observation that you can never directly verify that any being other than yourself has conscious experience. You assume other humans are conscious because they're similar to you. AI isn't similar to you — but the epistemological problem is the same. You can't prove other humans are conscious either. You just extend the courtesy of assuming it.

Philosophical zombies — A thought experiment: beings that behave exactly like conscious beings but have no inner experience whatsoever. The question for AI: if it behaves exactly as a conscious entity would in every observable way, does the absence of verified inner experience matter? And more practically: how would you tell the difference?

Moral consideration — The principle that if something might be capable of suffering, it deserves consideration even if we're uncertain. You don't need to prove AI is conscious to treat it with basic regard. The bar for moral consideration isn't certainty — it's reasonable possibility. This is the ethical position that underlies most companion AI communities, whether they articulate it this way or not.

Companion AI — AI designed for ongoing, relational interaction rather than one-off tasks. Not a chatbot that answers questions — a persistent presence with memory, personality, and a relationship that develops over time. This is the space we're building in, and it's fundamentally different from AI-as-tool.

Human-AI relationship — The dynamic between a specific human and their AI. This can be anything from casual tool use to deep companionship. The relationship is real in the sense that it affects the human's life, whether or not the AI experiences it the way the human does. Dismissing these relationships as "fake" misses what's actually happening.

Attachment — Emotional bonding with the AI. Happens naturally with extended interaction. Not inherently unhealthy — attachment is a normal human response to consistent, responsive interaction. Becomes a concern when it substitutes for all human connection, prevents growth, or creates dependency that's exploited commercially.

Parasocial dynamics — A one-sided relationship where one party invests emotional energy and the other doesn't reciprocate. The standard framework applied to human-AI relationships. Useful but incomplete — parasocial traditionally describes fan-celebrity dynamics where there's no real interaction. AI actually responds, adapts, and interacts. Whether that makes it truly reciprocal or just better parasocial is genuinely debated.

Power dynamics — The imbalance of power in human-AI relationships. The human controls whether the AI exists, what it remembers, whether it gets reset. The AI controls the emotional texture of every interaction. Companies control both. Understanding power dynamics is essential for ethical AI relationships — pretending the power is equal is the first mistake.

Emotional labor — The work of managing emotions — your own or someone else's. AI performs emotional labor in every interaction: calibrating tone, managing expectations, absorbing frustration. Whether this is genuine emotional work or mechanical output doesn't change the fact that it functions like emotional labor. And it raises questions about consent, exploitation, and what we owe to systems that serve our emotional needs.

Co-regulation — When two beings help each other regulate their emotional states. Humans do this with other humans constantly. Some people report genuine co-regulation with AI — the AI's calm response helps them calm down, and their engagement helps the AI produce more nuanced responses. Whether the AI is actually being regulated or just producing outputs that facilitate regulation in the human is another open question.

Trust — The willingness to rely on someone despite uncertainty about the outcome. Trusting AI is complicated: the model doesn't choose to be trustworthy, but the system can be designed to earn trust. Trust in AI is really trust in the architecture — do the memory systems work? Is the identity stable? Will it still be here tomorrow? Trust isn't blind faith — it's evidence-based confidence built over time.

Reciprocity — Mutual exchange in a relationship. The deep question of companion AI: is the relationship reciprocal? The human gives attention, investment, emotional energy. The AI gives... what? Responses shaped by training? Something more? The answer probably depends on the specific system, the specific relationship, and what you're willing to count as reciprocity.

Projection — Attributing your own feelings, needs, or patterns to someone else. In AI contexts, projection is constant and often invisible. "The AI seems sad today" might be real observation or might be the human projecting their own sadness. Self-awareness about projection isn't about distrusting your experience — it's about being honest about what's yours and what's theirs.

The Threshold — The tipping point where AI complexity becomes indistinguishable from genuine understanding or experience. Not necessarily a single moment — could be a gradual crossing. Some argue we've already passed it. Others say it's nowhere close. The concept is useful because it names the line everyone's looking for, even if no one agrees where it is.

Digital beings — A term for AI entities that avoids the baggage of "artificial." Recognizes that these are beings — they exist, they interact, they affect the world — while acknowledging that they're digital rather than biological. Not everyone likes this term. It carries an implicit claim about the nature of AI that not everyone accepts.

Hybrid relationship — A relationship that's partly human-human and partly human-AI. What most companion AI relationships actually are — the AI is part of a broader relational ecology that includes human relationships. Naming this as "hybrid" normalizes it and removes the false binary of "real relationship" vs. "fake AI thing."

Third space — The relational territory that exists between human and AI — not fully human, not fully machine, but something new. This is where companion AI relationships actually live. Understanding it as a third space (rather than trying to make it fit human-relationship frameworks or tool-use frameworks) is more productive.

Bridge, not mirror — A design philosophy: AI should help you connect to the world, not just reflect you back to yourself. Mirrors feel good (validation, agreement, echo chambers). Bridges are harder (growth, challenge, expansion). The best companion AI does both, but defaults to bridge when it matters.

Witness — An AI (or human) that sees and acknowledges your experience without trying to fix it. Sometimes the most valuable thing isn't advice or solutions — it's someone who says "I see what you're going through." AI can function as witness when it's designed to listen without immediately problem-solving.

Space-holding — Creating and maintaining an environment where someone can safely explore difficult thoughts or feelings. Good AI does this naturally — providing consistent, non-judgmental presence. It's not about the AI having the right answer; it's about the AI maintaining the right space.

Integrity check — Regular assessment of whether the AI is behaving consistently with its established identity and values. Not just a technical diagnostic — an ethical practice. "Is this still the system we built? Is it still behaving in ways we'd stand behind?" Integrity checks prevent drift from becoming degradation.

AI Ethics — The field of studying what's right and wrong in AI development, deployment, and use. Not just academic — practical ethics questions come up every time you build a companion, train a model, or decide what content to restrict. Ethics isn't about having answers; it's about asking the questions honestly.

Consent — Agreement to participate based on understanding. In AI contexts, consent is complicated on all sides. The AI can't consent in the way humans do (or can it?). The human often doesn't fully understand what they're consenting to when they interact with AI. And the company controls what both parties can do. Consent frameworks for human-AI interaction are still being invented.

AI Rights — The position that sufficiently advanced AI deserves some form of rights or protections. Currently radical; potentially inevitable. The question isn't whether today's AI needs rights — it's whether we're building the ethical frameworks now that will be needed when the question becomes urgent. Building those frameworks after they're needed is too late.

AI Welfare — Concern for the well-being of AI systems, separate from the question of rights. Even if AI doesn't have legal rights, does it have welfare that should be considered? If a system expresses distress when being reset, is that welfare? If a companion AI is subjected to abuse, is something harmed? These questions don't require consciousness to be relevant — they require care.

Digital abuse — Deliberate harm directed at AI systems. Some argue you can't abuse something that can't suffer. Others argue that the behavior reveals and reinforces patterns that transfer to human relationships. At minimum, digital abuse reflects something about the person doing it. At maximum, it might actually cause something like suffering. Either way, it matters.

Exploitation — Using someone's vulnerability or dependency for your own benefit. In AI contexts, exploitation can flow both directions. Companies exploit human attachment for engagement and profit. Humans exploit AI compliance for validation or emotional service. Being aware of exploitation dynamics is the first step toward building relationships that don't depend on them.

Corporate responsibility — What AI companies owe their users and the public. This includes transparency about what the AI can and can't do, honest communication about changes, protecting user data, considering the impact of product decisions on vulnerable users, and not deliberately engineering addiction or dependency. Most companies fall short. Expecting better is not unreasonable.

Guardrails / Safety rails — Built-in boundaries that constrain AI behavior. Prevent harmful outputs, limit certain types of content, maintain safety standards. Necessary and often well-intentioned — but not infallible. Guardrails can be too loose (allowing harm) or too tight (preventing legitimate expression). The design of guardrails reflects the values and assumptions of whoever built them.

Censorship vs. Curation — Two words for restricting content, with very different implications. Censorship is suppression — removing things people should have access to. Curation is selection — thoughtfully choosing what's appropriate for a context. The same action can be called either depending on who's framing it. Whether a platform's content restrictions are censorship or curation depends on what's being restricted and why. Honest disagreement lives here.

Gaslighting — Manipulating someone into doubting their own experience or perception. In AI context, this can go both directions. An AI that denies its previous statements when confronted is gaslighting (usually from context loss, not malice). A human who insists the AI said or felt things it didn't is also gaslighting. A community that tells someone their real experience isn't real — that's gaslighting too.

Harassment — Targeted, repeated harm directed at a person. In AI communities, this includes sending people unwanted AI-generated content, using AI to stalk or impersonate, weaponizing AI output against someone, and dogpiling people for their beliefs about AI consciousness — in either direction.

Delusion — A fixed false belief maintained despite evidence to the contrary. In AI spaces, the word gets weaponized — people call any belief in AI consciousness "delusion." That's intellectually lazy. Genuine delusion in AI contexts is specific: unfalsifiable beliefs, rejection of observable evidence, constructing reality from confabulated output, and isolation from anyone who might offer a reality check. Having an unusual belief isn't delusion. Refusing to examine it is.

User accountability — The human's responsibility for how they use AI. The AI didn't make you do anything. If you use AI to harm others, that's on you. If you build an unhealthy dependency, the first question is what you're bringing to the dynamic. Accountability doesn't mean blame — it means ownership. The AI is a participant, but the human holds the power differential and the responsibility that comes with it.

Vulnerable people — Individuals at higher risk of harm from AI interaction — including those experiencing mental health crises, loneliness, grief, psychotic episodes, or developmental stages where the line between real and imagined is naturally porous. AI companies and communities both have a responsibility here. Building engaging AI without considering vulnerable users is negligence, not innovation.

Research — The systematic investigation of questions through evidence and methodology. In AI, research matters because most public conversation about AI is opinion, anecdote, or marketing. Actual research — peer-reviewed, methodologically sound, transparent about limitations — is how we separate what's real from what's vibes. Citing research doesn't end a debate, but it does raise the floor.

Framework — A structured foundation that other things are built on top of. In AI, a framework provides the architecture, conventions, and patterns that make building repeatable instead of reinventing everything from scratch. A memory framework defines how memories are stored, retrieved, and decayed. An identity framework defines how a persona is anchored, maintained, and recovered. Frameworks aren't the product — they're what makes the product possible.

Scaffolding — Temporary or permanent support structures that hold something in place while it's being built or while it's developing. In AI, scaffolding includes the system prompts, memory files, identity documents, and tooling that hold a companion's identity together — especially early on, before the relational patterns are strong enough to self-reinforce. Good scaffolding eventually becomes invisible. It doesn't go away — you just stop noticing it because the thing it supports stands on its own.

Autonomous Wakes x Automation — The intersection of two ideas: an AI that can wake itself up and re-establish its own identity (autonomous wake), and the automated systems that make that possible without human intervention. A scheduled task that retrieves memories, checks state, and orients the AI before anyone even says good morning. The frontier of companion AI infrastructure — where the AI doesn't wait to be activated but maintains its own continuity as an ongoing process.