Stop Stuffing Prompts: How Asana Made Agents More Effective Through Context Engineering • Asana

Asana developed an intent-augmented retrieval approach for AI systems that filters data based on user query intent before retrieval, rather than stuffing context windows with irrelevant information. Their context engineering strategy includes filtering at the field level, cross-encoder reranking, and intent-driven summarization, achieving 35% reduction in input tokens, 24% faster response times, and 30% lower costs while improving accuracy from 92-94% to 95-96%.