Context Quality Engineering

Context quality engineering is the production discipline of selecting, compressing, deduplicating, and structuring only the decision-relevant information an LLM or agent needs.

Key points

• Datadog reports that model context windows have expanded from roughly 128,000 tokens to as high as two million tokens in some pricing tiers ^[src-037].
• Larger windows have shifted the constraint: the problem is often no longer raw capacity, but whether the inserted context actually helps the model decide ^[src-037].
• Datadog observed that average request tokens more than doubled for median customers and quadrupled for 90th-percentile power users year over year ^[src-037].
• As teams stuff conversation history, retrieved documents, tool outputs, state, and guardrails into prompts, noise and redundancy can bury critical details ^[src-037].
• The report states the new limiting factor clearly: context quality, not context volume, determines reliable production agents ^[src-037].
• Effective systems maintain retrieval quality, summarization, deduplication, compression, and information hierarchy so long-context models receive high-signal inputs ^[src-037].
• Google Cloud adds a governance-related quality constraint: policy text can be necessary, but too much policy in-context can distract the model from the task; hard guardrails can keep policy outside the reasoning window ^[src-043].
• [[Context-sharding]] keeps each agent focused on a smaller work slice with only the instructions and skills needed for that slice ^[src-043].
• Knowledge Catalog and Workspace Intelligence represent a productized version of context quality: organize data semantics and workflow context so agents receive trusted, relevant business context instead of unstructured dumps ^[src-044].

Related entities

• Datadog

Related concepts

• Context Engineering
• Context Rot
• Retrieval-Augmented Generation (RAG)
• LLM Knowledge Bases (Karpathy pattern)
• Prompt Caching for Agents
• Claude Code Token Economics
• Context Sharding
• Containment Over Constraint
• Enterprise Knowledge Graph
• Workspace Intelligence

Source references

• ^[src-037] Datadog — “State of AI Engineering” (2026-04-21)
• ^[src-043] Google Cloud Events — “Operationalize AI: A blueprint for managing enterprise agents at scale” (2026-04-24)
• ^[src-044] Thomas Kurian — “Welcome to Google Cloud Next ’26” (2026-04-22)