A framework called the 'work factory' is proposed for structuring AI-assisted knowledge work: a local folder containing three layers - a data layer (MCP-connected live sources like Jira, Confluence, Salesforce), a knowledge layer (curated domain context and run logs), and a skills layer (step-by-step instruction files that guide the AI through a specific task). The approach draws an analogy to software design patterns and software factories, arguing that intentional scoping and feedback loops make AI outputs more consistent and reliable than ad-hoc prompting. A detailed sales example shows how opportunity assessment and weekly pipeline reporting can be automated using skill files, frameworks, and templates connected via MCP servers to CRM, Confluence, and Google Drive. Guidance is given for getting started with one factory job and scaling to multiple factories over time.
