Data Becomes Context

In 30 seconds

AI does not need as much data as possible, but the right, current, authorised and traceable excerpt. Context thus becomes its own, governed product layer — with provenance, meaning and purpose limitation.

Why it matters

If a model is let loose uncontrolled on raw data, wrong answers, permission leaks and non-traceable statements arise. More data does not automatically improve quality.

Only once provenance, currency, authorisation and purpose are machine-readable does data become reliable context.

Editorial assessment“Context Engineering” is today often more important than prompt tuning: most bad answers stem from poor or unauthorised context, not from a weak model.

The core architecture questions

• Which sources may a given agent use for a given purpose?
• How current must the context be — and how is staleness detected?
• Where does a piece of information come from, and can this be traced back to the source?
• How are the users’ permissions carried over to context access?
• How are sensitive contents excluded from indexing?

Typical patterns

• Context as a Product: curated, versioned context products instead of ad-hoc access.
• Metadata & Lineage: provenance and transformation path of every piece of information.
• RAG & Knowledge Graphs: targeted retrieval instead of everything-in-the-prompt.
• Provenance: every statement is traceable back to its source.
• Authorisation & Purpose Limitation: access follows role and purpose, not convenience.

Common misconceptions & anti-patterns

• The entire data lake is indexed unfiltered.
• Users’ permissions are ignored in retrieval.
• Context without provenance — answers cannot be verified.
• Outdated contents are reused without labelling.

Practical review questions

1. Can we name the sources used for an answer?
2. Does context access respect the user’s permissions?
3. Is currency measured and outdated context marked?
4. Are sensitive data demonstrably excluded from the index?

Link to the end-to-end worked example

Which data sources may the agent use, how up to date are they — and where does a piece of information come from?

• Only approved sources (knowledge base, case history, master data) — no unfiltered access to everything.
• The permissions of the human handling the case are carried over to the context access, not bypassed.
• Every piece of information drawn upon can be traced back to its source (provenance), so the recommendation is verifiable.
• Outdated or blocked content is flagged or excluded from the index.

To the full worked example: service agent →

Link to the reference architecture

This mission forms the “Context, Knowledge & Memory” layer of the reference architecture. It supplies the material that lets agents work reliably in the first place — and ties closely into Identity (004) and Tools (003).

Primary sources

Change history

• 25.06.2026 — Editorial Structured according to the mission template.
• 24.06.2026 — Editorial (subject matter) Initial publication with primary sources.