Context Engineering: From Prompts to Corporate Multi-Agent Architecture
arXiv 2603.09619·medium signal
Formally defines context engineering (CE) as a standalone discipline for designing, structuring, and managing the full informational environment in which an AI agent operates — framing context as 'the agent's operating system.' Proposes five context quality criteria (relevance, sufficiency, isolation, economy, provenance) and extends the framework to multi-agent organizational architectures where context pipelines span teams of agents. A useful conceptual framework for practitioners where context management has become the dominant engineering challenge.