Context Retrieval
person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation
Context retrieval is the process by which an AI agent fetches relevant background information from external sources to inform its responses, enabling it to ground outputs in up-to-date or domain-specific knowledge rather than relying solely on its training data.
Usage in GTM Engineering at Anthropic
In Anthropic's go-to-market (GTM) engineering stack, context retrieval is a core component of CLAFTS, an AI email assistant powered by Claude. The system uses a Google Docs integration as the retrieval backend, pulling from internal documentation to supply the model with relevant context at inference time. As described in the talk: "for the context retrieval, we're going to use our uh integration with Google Docs" — How Anthropic uses Claude in GTM Engineering, 1:57.
The practical motivation for this architecture is that it allows sales representatives to respond more technically and accurately to customer inquiries without manually searching knowledge bases themselves. By automatically retrieving the right documentation and feeding it into the model's context, the assistant effectively closes the gap between a non-technical sender and the precise, detailed answer a customer may need. How Anthropic uses Claude in GTM Engineering, 1:57
Role in Agent Architecture
Context retrieval functions as a distinct layer in the agent pipeline — separate from generation — and is what enables the system to stay current with internal knowledge that would not exist in a pretrained model's weights. In this implementation, Google Docs serves as the live knowledge base, making the quality and organization of that documentation directly consequential to the assistant's output quality.