Skip to content
Back to Magazine
ai-operating-models Updated 2 min read

RAG 2.0: from retrieval to governed context

Key takeaways
  • → Fuentes gobernadas: ownership, permissions and versioning of each source.
  • → Contexto minimo: what the agent must know to decide confidently.
  • → Calidad operativa: evaluation criteria per source and per use case.
  • → Does each source have an owner and permissions?
Contents

Problem

Many companies say they have RAG because they connect a search engine to a model. But in practice, the context remains dirty, disordered, and without ownership.

The result is predictable: inconsistent answers, dubious decisions, and a false sense of control.

Thesis

RAG 2.0 is not just retrieval. It is governed context: sources, permissions, versioning, and quality criteria.

Callout — Without context governance, RAG only amplifies noise.

Framework

Three layers that turn RAG into a system:

  • Fuentes gobernadas: ownership, permissions and versioning of each source.
  • Contexto minimo: what the agent must know to decide confidently.
  • Calidad operativa: evaluation criteria per source and per use case.

Mini‑case: a team connected 12 sources to their RAG. The answers improved in volume but not in reliability. By reducing to 4 sources with clear ownership and quality criteria, precision rose and reversions dropped.

Anti‑example: thinking that more documents = more intelligence.

Posture: RAG without governance is not an advantage. It is operational risk.

Breathing: In practice, the problem is not the model. It is the dirtiness of the context.

Protocol (3 steps)

  1. Define ownership de fuentes: each source has a responsible party and clear permissions.
  2. Fija contexto minimo: for each use case, what context is mandatory.
  3. Evalua calidad por fuente: if a source fails two cycles, it is corrected or removed.
Signal Metric Threshold
Source quality % valid answers > 90%
Reversion % decisions reverted must fall cycle to cycle
Coverage % cases with minimal context > 95%
Quick checklist for governed RAG
  • Does each source have an owner and permissions?
  • Is minimal context defined per case?
  • Are quality criteria defined per source?

Next step

If your RAG today answers but doesn’t decide well, schedule a diagnosis at contact.


Translated from the Spanish original with AI assistance and reviewed for accuracy. Read the original in Spanish.

Prepare a meeting with this article

Decision

Decide what governance, ownership or cadence is missing before scaling AI.

Meeting

Executive committee, AI portfolio review, transformation steering.

Risk

Mistaking activity, pilots and tooling for real operating capability.

Agent prompt: map decision rights, KPIs, risks and the next operational move

ai-operating-models context-engineering operating-model
Cite this article

Berthelius, V. (2026). “RAG 2.0: from retrieval to governed context”. BRTHLS Magazine. https://www.brthls.com/magazine/rag-2-0-governed-context-en

Read next

  1. Editorial connection

    AI Search Operating System: Why Enterprise Search Is Becoming the OS for Internal AI

    The AI search operating system is the governed context layer internal AI runs on. Here is why enterprise search is becoming the OS for agents, and how to build it.

    Updated 4 min read

  2. Editorial connection

    Gemini 3.1 Pro: what actually improved (and what changes in operations)

    Google just shipped Gemini 3.1 Pro. Beyond launch noise, the useful signal is simple: the baseline for complex AI work moves up again.

    2 min read

  3. Editorial connection

    Context Architecture: why prompt engineering does not scale a business

    This is the critical angle of the Context Architecture pillar. Read the full pillar first, then use this piece as the operational critique.

    3 min read

Fractional CAIO · Free diagnostic

Is your company ready to operate with AI?

30 minutes. No pitch. An honest read on where you are and what to move first.

Book free diagnostic