(noun)/dɑrk ˈkɑn.tɛkst/
What your organization knows but never wrote down. A general AI language model (LLM) holds all the general knowledge but none of yours. That gap is your dark context.
Most of what a company knows is never recorded. It lives in people, and when someone quits, a piece of how things work leaves with them. Unwritten and unmeasured, it is what most companies operate on.
Similar institutional memory, know-how, silent knowledge, tacit knowledge, tribal knowledge
There is a mass you cannot see
In physics, dark matter is the unseen mass that holds the galaxies together, never seen directly, known only because everything visible moves around it. Every organization has its own: the knowledge no one wrote down, the decisions never recorded, the sense of how things are done that lives in heads and the space between meetings. It sits in no document, and it holds the whole place together.
This is the context AI is now hungry for, and the gap it cannot see.
AI just made invisible knowledge expensive
For a long time this was a slow leak. Someone left, a little know-how walked out, the work carried on, a cost no one had to name. Most organizations already measure the hard things, hours billed, uptime, cash, and almost none measure the soft ones: what gets said in a meeting, how a decision actually moved, who quietly knows the thing. AI now needs the soft side, and most companies have neither the language nor the instruments for it.
A model is pure capability with no idea how you work. Give it your dark context and it moves with you; give it nothing and you get answers that are confident, generic, and slightly wrong. A driver with a license and no map of your roads.
Two never captured, two out of reach
Dark context comes in four shapes. Two are capture failures: the knowledge was never expressed or never written. Two are access failures: it was written, but it cannot be reached.
Unarticulated
Silent knowledge. Nobody has put it into words yet. It lives in hands, habits, and the sense of how things are supposed to go.
Articulated
Said out loud, not written. Someone could tell you if you asked. Nobody asked. Nobody wrote it down.
Lost
Documented, but buried. No one finds it when they need it. Being written down is not the same as being reachable.
Fenced
Documented, but gated. Only certain people can reach it. The rest of the team operates without it, and no AI sees it either.
Capture failures and access failures call for different solutions. Both keep the knowledge invisible, and both leave the AI working from a gap.
Context is the asset no one is pricing yet
First companies bought labor, then attention: what you preferred, what you clicked, how you felt. The next thing of value a person produces is quieter and harder to see. It is context, the situated sense that makes a general system specific to your Tuesday, your customer, your craft.
That context is becoming the most valuable thing you make at work, and almost nobody treats it as something they own. A thing with no name has no price, and a thing with no price gets taken, or priced by someone else. Dark Context starts from the opposite posture: name it as yours and hold it, before it gets priced for you.
Take the knowledge carelessly and you lose the knower
When tacit knowledge is harvested without care, what gets lost is not the data. It is the knower. The role thins to a source of context, the judgment gets smoothed into clean prose, the room for doubt closes: there is no place to say I am not sure inside a training set. What is left reads well and means less.
There is a tragedy of the commons in this: a model can graze freely on the living expertise of the people who do the work. So the real question is not capture, it is incentive, how do you make it worth someone's while to be the person who knows that corner of the work best, and to keep it sharp. Answer that and the knowledge has a keeper; leave it unanswered and even the best capture quietly rots.
From a void to something you own
AI that works from your reality
The same tools that gave you generic answers start to sound like they know the place.
A map the whole team shares
A picture that lived in one head becomes one the group holds together, before any automation does.
Memory that outlives the person
The same artifact that briefs a new hire briefs an agent. The work stops walking out the door.
A clearer sense of your own shape
It shows what you need from others, the right hire, partner, or task to hand off, and opens moves you could not see before.
From silent to machine-runnable, one step at a time
Knowledge does not jump straight from a person's hands into a running system. It moves up in steps. Silent, then said out loud, then written, then teachable, then something a machine can run. The work happens at the bottom of the staircase, where a human still holds the frame and decides what is worth saying.
A proven way to get people aligned
Dark Context is built on MethodKit, a method that has helped governments, companies, and schools structure complex subjects and get on the same page for over a decade, in more than 120 countries. Getting a group to one shared picture of how things really work was always the hard part. It is also exactly the context an AI needs.
Sometimes the point is to keep it dark
Dark context usually gets framed two ways. As a risk: the knowledge that walks out the door when someone leaves. As an opportunity: the knowledge you finally capture and put to work. There is a third, and it matters as much as the other two.
Sometimes keeping something out of the record, away from the next hire, the vendor, the model, is not a gap to close. It is the feature. A journalist protects a source. A clinician holds what was said in confidence. The value is not in surfacing the knowledge, it is in deciding, on purpose, what stays unwritten and who never gets to see it. Never let a tool make that call for you.
A model that cannot see something cannot leak it. Sometimes the most valuable thing you can do with a piece of dark context is decide that it stays exactly where it is.
Where should we look next?
Point us at a field where unwritten knowledge runs the show. One line is the whole ask. It tells us where to dig next.
Who's behind it
Ola Möller
Has spent a career mapping how people see and talk about things: citizen photojournalism that set different realities side by side, art curation, and taxonomies for how a group talks about a subject, always by interviewing people about how they really work and the mental models underneath.
Andriy Zhukov
Has spent a long time trying to make machines more human: building video games as a kid, and trying to build AI in high school, long before it got easy. The throughline is one stubborn question. How do you get a system to meet a person where they actually are?
darkcontext:~$want to keep thinking about this?
Leave an email if you want to follow the idea with us. No pitch, just the occasional note as we map where dark context shows up and what helps bring it to the surface.