How it works

Better answers from your AI, and what it needs from you first.

Three workshop formats on one method, and a monthly review that keeps it current.

Every AI setup has three parts

Two are rented. One is yours.

  1. The app — the AI program you use.
  2. The model — you rent this.
  3. The context — your specific knowledge and your setting. This one is yours.

Engineers call the app the harness.

Personal AGI →

When you switch providers, your context comes with you, because it never belonged to the provider.

Rent the model. Own the context.

dark context noun

What your organisation knows but never wrote down. The biggest part, and the part a general AI model cannot see.

What is dark context →

Most of it sits on the lower steps: unsaid, said once in a meeting, or written and filed where your AI cannot reach it.

Think of the colleague who also keeps the printers running. That work is in no job description.

Your AI can only see the last step.

Two ladders, one map

Two things decide which format fits: how much of you your AI can see today, and where your memory sits.

One rule runs under both: each step up in what you ask of your AI needs a step across in what it can see.

What your AI can see

Five stages, from Lights off to Kept lit. Find the row that sounds like your team today.

At your placeWhat usually goes wrongNext step
Lights off · answers: silentPhone, paper and meeting in person. Someone else does the books, digitally. It works just fine.All it knows is silent, in hands and habits. What AI use there is goes underground and takes the record with it.The session, for the leadership team.
Low light · answers: genericEmail, online docs, templates. Meetings online, unrecorded. Everyone has a seat; most go quiet within a month.Little is written, less reaches the AI. It answers as a stranger, and the checking never stops.The session.
One lamp · answers: someone'sNotes transcribed, documents digitised. One or two people do a team's work with it; the rest have heard the stories.Knowledge starts to get written, by the few for the few. Their speed rests on what only they know, so it never spreads.The mapping.
Corners lit · answers: on pointContext files written for it, rules for what it reads. "Did you check it?" becomes "what did it miss?"The first pilot worked. Six months on, the files it reads are stale and nobody owns them.The build, then the monthly review.
Kept lit · answers: yoursThe company reimagined around its memory. People set the direction and step in when something is off.The AI acts only on what is connected. Whatever stayed in heads never reaches it.Keep it current.
Not sure? Take the knowledge audit: twelve statements, two minutes. →

Where your memory sits

  1. Unsaid: in a head. For example, why you priced it that way.
  2. Verbalized: said once, in a meeting. For example, what you decided over lunch.
  3. Documented: written, out of your AI's reach. For example, last year's offer, in a folder.
  4. Connected: your AI reads it and acts on it. For example, examples and guidelines your AI can use.

Each of the three formats takes your memory one step up.

  1. Say it · the session What the room knows, said out loud, in its own words.
  2. Write it down · the mapping One corner of your memory written down by the people who hold it, with a cadence to keep it current.
  3. Connect it · the build Context files your AI reads and acts on, and a first prototype on a real task.
  4. Keep it current · the review Once a month we go through what your AI got wrong with you, and fix the context behind it.
The longer ladder: the 7 context levels →

Why a room

  1. Everyone talks about how to prompt AI. We work on how you prompt people. Knowledge often stays dark because people do not talk about it, or nobody asks.
  2. Most of your context is not in documents. Nobody fills in a hundred-question form about that, and nobody should. A good hour of talk brings up more than a form does in a month.
  3. A room gives people permission to share.
  4. A recording, an interview or an upload keeps what you already said. The room makes you say it.
  5. Anyone can make a file of the same shape. The room is what fills it. We make keys. The map is yours.
Conversational maps →

You leave with a file, not a feeling.

Each of the three formats ends with something you keep. The review keeps it current.

  • The session: this-room.txt, everything the room wrote down, in one file.
  • The mapping: a context map of your organisation, with the dark corners marked.
  • The build: context files your AI can use from day one, and a working first prototype.
  • The review: each month, a fix to the context behind what your AI got wrong.

The same file briefs a new hire and an agent.

And what should stay dark

Not everything belongs in the record. You decide what goes in and what stays yours.

You need to know which secrets you have before you can protect them.

A model cannot leak something it cannot see.

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.

Never let a tool make that call for you.

Sometimes the point is to keep it dark →

Why the review exists

Written down does not mean up to date.

That is why the review exists. It follows any of the three formats, for as long as people and priorities keep changing.

Find your step, then talk to us.

Tell us where your team's memory sits today.