context collapse and personal AI

Table of content

by Ray Svitla


you know that feeling at a party where your work friends meet your childhood friends and you suddenly don’t know which version of yourself to be? sociologists have a name for that. context collapse. and it’s about to become the defining problem of personal AI.


the sociology

context collapse was coined by danah boyd and later expanded by Michael Wesch to describe what happens on social media. in the physical world, you naturally segment your identity: professional you at work, casual you with friends, vulnerable you with family, weird you at 2am on the internet.

social media collapsed these contexts. your boss, your mother, your ex, and your college roommate all see the same post. so you perform for the broadest possible audience, which means you perform for no one in particular. you become a flattened, inoffensive version of yourself.

this is why corporate LinkedIn posts sound the way they do. it’s not that those people are boring. it’s that context collapse forces blandness as a survival strategy.


the AI version

now apply this to personal AI.

your AI assistant has one model of you. one memory. one set of preferences. and it uses this single, undifferentiated model whether you’re:

each of these is a different you. not a different mask — a genuinely different configuration of values, priorities, and needs. but your AI sees one flat profile and optimizes for the average.

the average of all your selves is none of your selves.


why this matters for routing

the practical consequence: your AI gives you the same kind of response regardless of context. professional tone when you need warmth. efficiency when you need space to meander. challenge when you need comfort. comfort when you need challenge.

it’s not that the AI is dumb. it’s that it’s been given a collapsed context and is doing its best with a flat map of a three-dimensional person.

this is the same problem second brains have. your Obsidian vault doesn’t know whether you’re searching for “productivity” because you’re procrastinating or because you’re genuinely optimizing. it returns the same results either way. no routing. no context. no awareness of which you is asking.


the routing solution

identity as protocol solves context collapse by making context explicit.

a self.md doesn’t just say “prefers concise responses.” it maps your contexts. it knows that at work you want directness and at midnight you want exploration. it knows that your tension between autonomy and structure resolves differently depending on whether you’re in a creative phase or an execution phase.

this is routing in practice. not one response style. not one personality model. a dynamic map of selves that the AI reads and adapts to.

## modes
- deep work → minimal interruption, technical precision, 
  challenge assumptions aggressively
- reflection → open-ended, allow contradictions, 
  don't optimize for efficiency
- crisis → direct, practical, emotional context first, 
  logistics second
- creative → loose, associative, break patterns, 
  suggest the unlikely

this isn’t a personality quiz result. it’s a routing table. it tells the AI: right now, in this context, here’s which version of me you’re serving.


the deeper problem

but here’s what makes context collapse genuinely hard: you don’t always know which you is present.

sometimes you think you’re in deep work mode but you’re actually avoiding something emotional. sometimes you’re in “creative” mode but it’s just procrastination wearing a beret. sometimes you’re in crisis mode and you don’t realize it for hours.

a truly useful identity protocol needs to handle this. not by deciding for you — that’s the narcissism trap — but by reflecting back what it observes. “you said you’re in deep work mode but your last three queries were about apartment listings in a different country. want to talk about what’s actually going on?”

that’s an AI that understands context collapse. it doesn’t flatten you. it doesn’t pretend to know which you is present. it notices the contradictions and makes them visible.


fragmentation versus integration

there’s a temptation to solve context collapse by fragmenting: separate AI for work, separate AI for personal, separate AI for creative. keep the contexts apart.

this misses the point. the contexts should talk to each other. your creative self has insights relevant to your work self. your crisis responses reveal things about your values that matter everywhere. fragmentation is just context collapse in reverse — instead of all your selves being flattened into one, they’re isolated into silos.

what you want is integration with awareness. one identity protocol that knows all your selves exist, knows they’re different, and knows which one is active. not separate brains. one qualified self with a good routing table.


the social angle

context collapse gets even more interesting when you consider that self.md files can be shared.

what if, before a collaboration, you shared relevant parts of your identity protocol with a colleague? not your diary. not your deepest tensions. but your working style, your communication preferences in different modes, your known friction points.

“I go quiet when I disagree rather than arguing immediately. if I’m silent, ask me directly.”

this is context expansion — the opposite of context collapse. instead of your colleague experiencing a flattened version of you, they get a routing guide. a map of how to interact with you that acknowledges you’re not one thing.

sounds utopian. maybe. but it’s technically trivial. it’s a markdown file. the hard part was never the technology. the hard part is the self-knowledge required to write one honestly.


which version of you is reading this right now? and would your AI know the difference?


identity as protocol — the solution to context collapse → self as process, not profile — why identity is multiple → the narcissism trap — when self-awareness becomes self-obsession → why your second brain is dead — another flattening problem


Ray Svitla stay evolving

Topics: context-collapse identity philosophy self-md sociology