What is authored context?

The Ingestion Distinction

Most AI systems build their knowledge base by ingesting automatically — optimising for coverage, pulling from whatever exists. The assumption is breadth: more in the system means more to retrieve from.

Authored context works differently at the ingestion stage. Knowledge isn't pulled in because it exists — it's put in because someone decided it should be there, in that form, with that reasoning attached. A human judgement is applied before the content enters the system: is this accurate? Is this the right level of specificity? Does the reasoning behind it need capturing alongside it?

Authored context optimises for intent and explainability rather than coverage. Less may be in the system, but what's there was put there deliberately.

That judgement is the difference. Not retrieval — ingestion.

What it Includes that Automatic Ingestion Misses

Automatic ingestion captures what has been written down. It doesn't capture the reasoning behind decisions that were made in meetings and stayed there. It doesn't capture the constraint that everyone on the team knows but nobody formalised. It doesn't attach the why to the what.

Authored context captures those things — because authoring is an active process. A human, working with an agent or independently, surfaces the knowledge that matters, gives it the right structure, and records the reasoning that makes it usable rather than just retrievable.

The result is a knowledge base that reflects what the organisation has explicitly committed to operationalising — not just what it has happened to write down.

What Authored Context is not

It is not a system prompt. A system prompt is written once, usually quickly, to get an agent working. It reflects what the author thought of at the time. Authored context is built over time, maintained as things change, and structured so that an agent can navigate it rather than receive it as a block of instructions.

It is not retrieved context. Retrieval finds relevant content from a corpus at inference time. Authored context is the curation discipline that determines what's in the corpus in the first place — and in what form, with what reasoning attached.

It is not telemetry or logging. Logs record what an agent did after it acted. Authored context records what the agent was given before it acted. These are different records serving different purposes.

Why the Distinction Matters

When an agent makes a decision, two questions can be asked about it. The first is what did it do — logs answer that. The second is what was it working from — authored context answers that, if it exists.

Without authored context, the second question can only be answered retrospectively, by reconstructing the agent's operating state from whatever evidence remains. With it, the answer is retrievable: the knowledge the agent drew on was curated before it acted, structured so it could be navigated, and recorded so it could be explained.

That's the governance property authored context provides. Not through monitoring or logging — through the discipline of curation before the agent acts.