The Context Owners manifesto
AI does the work.
Humans own the context.
AI can now draft almost any deliverable. What it can't do is decide what it should know. That job belongs to people, and it is becoming the centre of knowledge work. The people who do it are Context Owners.
AI drafts the deliverables now
There's very little stopping AI from drafting almost any deliverable: decks, models, documents, emails. Each new model makes that more true.
The catch: AI rarely gets it right on the first pass. Worse, it can be convincingly wrong, or so long that checking it takes longer than writing it would have.
Context is the new bottleneck
Models keep getting smarter. Their context windows don't keep growing: frontier windows have sat near 1M tokens for a year, because attention cost grows quadratically with length.
As intelligence stops being the constraint, context quality decides the result.
Context dilution
Too much information, too little signal.
Context pollution
Wrong or outdated information steers the answer.
Context rot
Models get less reliable as input grows, even on simple tasks.
Fix it before, not after
There are three ways to improve what AI produces:
- let it check its own work,
- make human review easier,
- or give it the right context.
The first two fix the output after it exists. The third fixes it before. That's where the biggest gain is.
Golden context
Golden context is the validated, current and governed business knowledge an organisation allows its AI to rely on, so that what AI produces can be accurate, relevant and complete.
Golden data controls what systems know. Golden context controls what AI knows.
Golden data follows deterministic rules, so it can be automated. Golden context is open-ended: never complete, always evolving, and only as good as its fit to the task.
Golden context needs an owner
Because golden context is open-ended, deciding what belongs in it takes human judgment. People know what matters, what was decided and why, and what never made it into a document. And someone has to answer for it.
Golden data has data owners.
Golden context needs Context Owners.
Curate
Choose the sources that matter, and leave out the ones that would dilute or pollute.
Validate
Rule on every change AI proposes. AI checks, a human approves.
Maintain
Keep the context current as the work moves, and catch what is drifting.
The human world model remains unmatched. That makes us the Context Owners.
The work changes
The hours that went into drafting go into three things: thinking the problem through, keeping the context current, and validating what comes back.
That work is more interesting than building decks from scratch. It's where the thinking happens.
Augment, don't automate
Chasing one-shot, end-to-end automation is an unreasonable goal for most business work, and it hides the real reward: AI that makes people better at their work.
AI is stochastic. Its output carries noise, the objective drifts as the work moves, and it can misread a clear instruction. Left alone, it lands far from where you meant to go.
AI shouldn't think for you. It should help you think better, and help you turn that thinking into action.
Principles
The right context, not more context.
Golden context is an asset, not a by-product. It outlives the project.
Every golden context has a named owner.
AI checks. A human approves.
Keep it current. Stale context is polluted context.
Bring your own AI. The context should outlast any one model.
Not more context. The right context.
Give AI the smallest trusted set of knowledge it needs to do the work well.